Closes 148/149 — the accept gate now lands code instead of just moving a row.
Driver gains createPullRequest / getPullRequest / mergePullRequest, verified
against the real Gitea API shapes rather than guessed: MergePullRequestOption
requires a capitalised `Do`, and supports `head_commit_id`.
That `head_commit_id` is the point. A human approved a SPECIFIC commit; if
someone pushed after that, merging anyway lands code nobody reviewed. So the
reviewed sha travels with the request and is checked twice — once by us for a
clear MergeStaleError naming both commits, and again by the forge, which closes
the race between our read and the merge. When it trips, the message says what
happened and what to do: re-review the current head, then accept again.
Accept merges BEFORE it transitions. If the forge refuses — conflicts, failing
checks, a moved branch — the task stays in awaiting_review and no event is
written. Reporting `merged` while main does not have the code is the single
worst thing this gate could do, and there is a test asserting the transition
never happens on a failed merge.
Failure modes are answers, not stack traces: no PR attached is approval-only and
succeeds; a repo with no forge connection explains that mcpctl cannot merge it;
an already-merged PR is treated as success so a retry after a timeout is safe.
owner/repo comes from `sourceFullName` when discovery recorded it, falling back
to parsing https, ssh:// and scp-style clone URLs, and returning null rather
than guessing when it cannot tell.
The merger is injected and optional, so an mcpd without a git provider still
runs and accept degrades to approval-only — which is also what the help text
now says, restored after the earlier commit that removed a claim we could not
back up.
Tests: 7 driver PR tests (Do field, head pinning, stale refusal, idempotent
re-merge, 405 → refusal), 10 merger tests incl. URL parsing, 7 accept-gate tests.
Local: build clean, workspace 2580 passed (+23), lint unchanged at 794 errors.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01N4wNHWf7xSwnZCWpJcyv9p
`mcpctl tasks accept --help` said "fires the merge". Nothing merges. There is
no createPullRequest or mergePullRequest anywhere in the codebase — the
git-provider client is task 148 and is not built.
Accept records human approval and moves the task to `merged`. Until 148/149
land, `merged` means "a human approved this", not "this is in main", and the
help text and service comments now say so. A user who accepted a task expecting
a PR to land would have been silently wrong.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Closes 167. The five new resources fell through to the generic key-value dump,
against the project rule that describe shows clean sections and tables. More to
the point, two things built earlier had no way to reach a human at all: the
roadmap endpoint had no client, and the Epic reconciliation was only visible as
raw JSON.
`describe epic` puts the delivery mismatch directly under the completeness
percentage it contradicts, because reading either one alone is the bug. Rendered
against a live mcpd:
Completeness: [....................] 0% (0/3)
...
Delivery: MISMATCH between what is claimed and what shipped
Tasks: 1 merged, 0 outstanding, 0 cancelled
Delivered, not claimed:
- wheels (1 merged, 0 outstanding)
No roadmap node owns it:
- engine, feel
plus the criteria checklist, accepted tradeoffs, and the roadmap with per-node
completeness and READY flags.
`describe task` derives the gates a task still owes from its risk and recorded
verdicts ("awaiting security_review, sre_review" vs "all passed"), and shows the
legal next transitions, PR linkage, verdicts and full history — so the answer to
"why won't this move" is on screen rather than requiring a failed attempt.
`describe repo` / `describe gitprovider` show the credential REFERENCE only.
The resolved value exists solely on the session-provision path and must never
reach a describe.
The roadmap, verdict and event fetches are all best-effort: an older mcpd 404s
and the resource still renders, rather than the whole describe failing over an
enrichment.
One bug found by looking at the output rather than the types: the Git section
opened on `!== undefined`, but JSON hands back null, so every task without a PR
printed an empty `Git:` header.
Local: build clean, workspace 2557 passed, lint unchanged at 794 errors.
Verified live against a local mcpd for epic, epictask and task.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01N4wNHWf7xSwnZCWpJcyv9p
Closes 175. Epic.doneCriteria and EpicTask.doneCriteria are both JSON blobs
joined only by a free-form id, and nothing checked that a node's ids existed on
its parent Epic. A typo produced a criterion belonging to nobody.
The reconciliation added in 170 surfaces those as `unowned`, which is honest but
late: the mistake was made at write time and is far cheaper to refuse there. Now
create and update both reject unknown ids, naming every one of them and listing
what the Epic actually has — or saying plainly that the Epic has no criteria yet,
which is the likelier cause when a whole roadmap comes back empty.
Enforced in the service, not the schema: production provisions with
`prisma db push`, so a constraint would never exist where it matters.
A node with no criteria of its own is still fine — not every roadmap node claims
one — and an update that doesn't touch criteria doesn't pay for the check.
This deliberately stops short of the EpicCriterion table with a real FK. That
remains the stronger answer if criteria ever need to be addressed independently,
but it costs a schema change plus migrating everything currently in the JSON
blobs, and the invariant closes the actual failure mode today.
Also files 177: agent chat --project reaches project prompts and MCP tools while
checking only run:agents:<name>. Raised by the agent that built membership,
which correctly refused to invent the policy mid-task. Pre-existing behaviour for
a legacy pin, but membership widens the blast radius considerably — it should be
a deliberate decision with a test.
Local: build clean, workspace 2557 passed (+7), lint unchanged at 794 errors.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01N4wNHWf7xSwnZCWpJcyv9p
`git add -A` swept .claude/worktrees/ into the previous commit as embedded git
repositories. They are transient checkouts created for parallel agent runs and
have no business in the tree — removed from the index and gitignored.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Merges feat/agent-project-membership (task 166/152). An agent can finally join
many projects: membership routes, per-membership role and optional brain
override, chat resolving context from the membership, and CLI attach-agent /
detach-agent / get memberships. The legacy Agent.projectId fallback is kept for
this release as the schema comment promises.
Conflict resolved in main.ts: that branch added its RBAC case to
mapUrlToPermission inline, while this branch had extracted the function to
middleware/url-permission.ts so it could be tested. Kept the extraction and
ported the case into it — joining a team is `edit:agents:<name>`, not
`create:agents`, which is the authority to mint a whole agent.
Also fixes something the extraction makes possible. security.test.ts contained a
hand-copied reproduction of mapUrlToPermission, because main.ts self-invokes
main() and nothing in it was importable. A security test asserting against a
copy stops testing the system the moment the real one changes, and this copy had
already drifted. It now imports the real function; all 18 assertions still pass,
against the code that actually runs.
Local: build clean, workspace 2550 passed, lint unchanged at 794 errors.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01N4wNHWf7xSwnZCWpJcyv9p
The AgentProject junction has existed since the data-model commit and nothing
read it. `grep -c agentProject src/mcpd/src/services/chat.service.ts` returned
0; chat resolved an agent's project from the legacy single `Agent.projectId`
scalar. So the schema claimed "an agent is a person who joins zero, one, or
many projects" (docs/agentic-teams.md §4) while the running system still gave
every agent exactly one project. A schema that lies about behaviour is worse
than a missing feature: it reads as done, so nobody builds it.
This makes the junction load-bearing.
Membership is now a real surface (task 152):
GET /api/v1/agents/:name/projects memberships
POST /api/v1/agents/:name/projects/:project join, body {role?, llm?}
DELETE /api/v1/agents/:name/projects/:project leave
GET /api/v1/projects/:name/memberships the roster, with roles
`role` is TEXT (developer|qa|security|sre in v1) and is validated in the
service layer, not the database — production provisions with `prisma db push`,
so a CHECK constraint written into a migration would never run there. Adding a
fifth role stays a one-line change.
RBAC: joining/leaving is `edit:agents:<name>`, not the `create:agents` the
default method→action mapping would have produced. Minting a whole new agent
and adding an existing one to a team are not the same authority; this mirrors
project↔server attach/detach already being `edit:projects`. Reading a roster is
`view:` on the side you asked from.
Chat honours it. `prepareContext` resolves the project from the membership when
the caller names one, and the resolution runs *before* the Llm lookup so a
per-membership `llmId` can replace the agent's pinned brain — and therefore its
whole failover pool — rather than patching the dispatch list afterwards. That
override is why the brain-ownership question (§15 Q1) can still go either way
without a migration: today the answer is per membership.
Naming a project the agent has not joined is an error that names the fix, not a
silent fallback. Answering from a *different* project's tools and prompts than
the one asked for is the failure mode worth being loud about.
The legacy scalar keeps working this release, as the schema comment promises:
`--project` accepts a membership *or* a legacy pin naming that same project,
bare `mcpctl chat <agent>` is unchanged, and `GET /projects/:p/agents` returns
the union of members and legacy-pinned agents so nothing disappears from a
project mid-migration (mcplocal's agents plugin reads that endpoint).
CLI: `mcpctl chat <agent> --project <p>` (previously rejected as "not both"),
plus `attach-agent` / `detach-agent` alongside the existing attach-server pair,
`get agents --project`, and `get memberships` with ROLE and BRAIN columns.
Completions come from the generator, not hand-edits.
Threads stay scoped to the agent, not to (agent, project) — §4 wants the
latter, but that needs a schema change and a story for what happens to those
threads when a membership is revoked. Deliberately not decided here.
Tests: 51 new (43 mcpd, 8 CLI). Full suite 2514 pass. Lint unchanged at
794 errors / 77 warnings.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The guard test caught it, which is its job — the generator derives flags from
the commander program, so any new option makes the committed files stale.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Merges the Epic reconciliation (task 170) and closes the gap that made it inert
from the terminal.
`mcpctl create epictask` had no `--criterion` flag while `create epic` did, so a
roadmap built from the CLI had no criteria on its nodes. The reconciliation
attributes a criterion to a node by id, so every criterion would have reported
as `unowned` and both mismatch lists would always have been empty — the feature
would have looked like it worked and detected nothing.
Files two follow-ups the reconciliation surfaced:
- 175: criteria and roadmap nodes are linked only by a free-form id convention
that nothing enforces. EpicTaskService never validates ids against the parent
Epic, ids can repeat across nodes, and a node can carry an id the Epic never
had. The reconciliation reports those as `unowned` rather than guessing an
owner, which is right, but the convention should become an invariant before
more is built on it.
- 176: `get epic -o yaml` now emits computed `completeness`/`reconciliation`,
which are not settable. apply tolerates them, but the round-trip rule says get
output is apply input.
Local: build clean, workspace 2499 passed, lint unchanged at 794 errors.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01N4wNHWf7xSwnZCWpJcyv9p
An Epic could report 0% while every Task under it was merged. That is not a
cosmetic gap — it is exactly the blind spot the Epic exists to remove
(docs/agentic-teams.md §1b: "we completed all the tasks and got a stick with
wheels instead of a car"). An Epic that cannot see its own delivered work
cannot detect that, because `completeness()` reads only `doneCriteria[].met`
and nothing flips those except a human hitting the criteria endpoint.
So read the same question a second time, from the delivery side, and report
where the two readings disagree — in both directions:
- deliveredButUnclaimed — unmet criteria whose owning roadmap node has
merged work. Somebody delivered and nobody ticked the box.
- claimedButUndelivered — criteria marked met with nothing merged
underneath. The box is ticked and nothing shipped. The dangerous
direction, and the one that is easy to forget to look for.
- unowned — criteria no roadmap node claims at all. They cannot appear in
either list above, so without a bucket of their own they would vanish
inside a clean-looking reconciliation.
Nothing is written. `met` stays exactly as a human or agent set it: deriving
it from task state would assert that finishing the tasks means delivering the
vision, which is the assumption under audit. The mismatch is reported, not
silently resolved.
Attribution is by criterion id, because that is the only link the schema
offers — EpicTask.doneCriteria is documented as "the decomposed subset of the
Epic's doneCriteria this node owns" and carries the same ids, but nothing
enforces it. Rather than guess an owner by name similarity or ordering, the
convention is treated as a claim and its failures are surfaced as `unowned`.
A foreign key from criterion to epictask would end the ambiguity; this reports
the ambiguity instead of hiding it.
Judgement calls, made explicit so they can be argued with:
- `merged` alone counts as delivered. `done` only leaves the developer's
loop and `awaiting_review` still needs a human; `reopened` means a merge
did not hold, so it counts as outstanding.
- `cancelled` is neither delivered nor outstanding. Counting it as owed
would leave a node permanently undelivered because someone abandoned a
task; counting it as delivered is a lie. It is reported on its own.
- a node with no tasks is undelivered, not exempt — a met criterion over an
undecomposed node is still a claim with nothing behind it. The empty task
rollup is what distinguishes it from "built but not merged".
- an epictask's own `status` is another claim, not evidence, so it is
carried for context and never counted as delivery.
Surfaced on GET /api/v1/epics/:id next to the completeness it contradicts —
a mismatch is only useful where the misleading percentage is already on
screen — and standalone at GET /api/v1/epics/:id/reconciliation for callers
that do not want a 200KB Epic body with it. Deliberately kept off the list
route, which would pay a query per roadmap node per Epic.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Closes 134/164. The five agentic-teams resources had RBAC slots but the
non-CRUD verbs had no cases, so they fell through to the method default and
every one of them read as `create:<resource>`. That is wrong in both
directions: it let anyone who may CREATE a task also ACCEPT one — the human gate
that fires a merge — and it refused a reviewer who may drive tasks but not
create them.
Corrected, following the verb the codebase already uses for "use this resource
to do work" (llms/infer, agents/chat, projects/chat):
- POST /tasks/:id/{transition,accept,reject,verdicts} → run:tasks:<name>
- {GET,POST} /gitproviders/:id/{repos,import,probe} → run:gitproviders:<name>.
Discovery calls OUT to the forge with its stored token, so holding
view:gitproviders should let you see that a forge is registered, not use its
credential.
- POST /epics/:id/{status,tradeoffs,criteria} → edit:epics:<name>. These rewrite
the vision record.
- GET /epics/:id/roadmap → view:epics:<name>
- {POST,DELETE} /epictasks/:id/edges → edit:epictasks:<name>
mapUrlToPermission moved from main.ts to middleware/url-permission.ts. main.ts
calls main() at module load, so nothing in it is importable from a test — an
authorization mapping that cannot be tested is an authorization mapping nobody
has checked, which is exactly how these verbs shipped mis-scoped.
15 tests now assert the mapping rather than the mere presence of the strings,
including that a bare create grant does NOT cover the gated verbs, that work
items and the inference queue never resolve to the same resource for any method,
and that the pre-existing mappings (llm infer, agent chat, operations, the
personalities/skills piggybacks) still hold.
Local: build clean, workspace 2478 passed (+15), lint back to the repo's
baseline 794 errors after making the five new match guards explicit.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01N4wNHWf7xSwnZCWpJcyv9p
The tasks/repos command groups added with the CLI work sit at the root of
`mcpctl --help`, but both only function inside a project context. Beyond the
clutter, it teaches the wrong data model: a task reads as a peer of servers and
secrets rather than a leaf of the epic tree.
Convention recorded as the `scope-commands-to-their-context` prompt in the
mcpctl-development project. Note `attach-server`/`detach-server` are already
hidden, so the pattern exists and was simply not applied consistently.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The audit found that none of the new resources were in BACKUP_KINDS, so
git-backup never captured them and a restore would lose every epic, roadmap
node, repo and forge registration. Closes 135/165.
Backed up: gitprovider, repo, epic, epictask — configuration and the vision
layer. NOT tasks: a work item carries in-flight state (lifecycle position,
verdicts, PR links) and restoring a stale one from git would resurrect work that
has since moved on. That is the same reasoning that keeps McpInstance out of the
backup, and there is a test asserting it stays out.
Two properties the tests pin, because both are the kind of thing that fails
silently:
- **Credentials serialize as a REFERENCE, never a value.** The backup repo is a
plaintext git checkout; emitting a resolved token would export every forge
credential to disk in cleartext. credSecretId/credSecretKey are internal
fields and become `credRef: {name, key}`; restore re-resolves against whatever
the secret backend holds now.
- **Relations serialize by name, not foreign key.** A restore into a different
database has different cuids, so a backed-up projectId would dangle.
Restore ordering follows dependencies: a forge before the repos that reference
it, an epic before its roadmap nodes.
Two deliberate refusals on the restore path: an Epic's `status` is never written
by a restore, so a stale backup cannot silently mark it delivered — that
terminal state is only reachable through setStatus and its rules; and a
GitProvider is not probed on restore, because the forge may legitimately be
unreachable while a backup replays and failing the whole restore for that is
worse than a row that needs a later `probe`.
The pre-existing APPLY_ORDER length guard caught this change, which is exactly
what it is for; updated 9 → 13 and tied it to BACKUP_KINDS.length so the two
cannot drift apart again.
Also marks 153 done — `mcpctl tasks accept/reject` shipped with the CLI work; the
generated graph had it blocked behind the git-provider merge, but the gate itself
does not need it.
Local: build clean, workspace 2463 passed (+6), lint unchanged at 794 errors.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01N4wNHWf7xSwnZCWpJcyv9p
Found while auditing what this branch is missing: skills are serialized into the
backup repo and enqueued on every mutation, but `importResource` and
`deleteResource` in main.ts have no `skill` case — so a git restore silently
drops every skill. SkillService has had upsertByName/deleteByName the whole
time; the switch arms were simply never added. Pre-existing, not introduced
here.
Also records ten tasks (164-173) for gaps found by auditing the branch against
the codebase rather than against the plan:
- 164 RBAC enforcement tests. Five resources and a namespace split shipped with
no test proving the gating works. Also flags that the non-CRUD verbs
(transition/accept/reject, import/probe) currently fall through to the default
method mapping and are probably mis-scoped.
- 165 Backup coverage for the agentic-teams resources — same class of silent
data loss as the skill bug above, for five more kinds.
- 166 AgentProject is inert: chat still resolves context from the legacy
Agent.projectId scalar, so "an agent joins many projects" — a core idea of the
design — is not actually usable yet.
- 167 describe formatters (currently a generic key-value dump, against the
project rule) including a client for the roadmap endpoint, which has no caller.
- 168 Docs, including the breaking `mcpctl get tasks` rename that exists only in
a commit message today.
- 169 Smoke tests for the forge flow.
- 170 Epic completeness ignores delivered work: an Epic can read 0% while every
task under it is merged, which is precisely the blind spot the Epic exists to
remove.
- 171 `mcpctl chat mcpctl` has never been run end to end — the tool layer is
verified, the LLM-picks-a-tool loop and the mcplocal token hop are not.
- 172 Nothing on this branch is live; everything was verified against a local
mcpd on the test database.
- 173 Web UI + Pulumi kinds (cosmetic).
Local: build clean, workspace 2457 passed, lint unchanged at 794 errors.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01N4wNHWf7xSwnZCWpJcyv9p
Closes two stated project rules this work had been violating.
**`get -o yaml` must round-trip through `apply -f`.** The five new resources
were creatable and listable but absent from apply, so what `get` emitted was not
valid apply input. They are now applied in dependency order — gitproviders →
repos → epics → epictasks → tasks — since each references the previous by name,
and each block is independently fault-tolerant so one bad entry does not abort
the rest of a manifest. `edit` accepts them too.
`state` is deliberately NOT in the task spec. Lifecycle moves go through
`mcpctl tasks advance`, which enforces the transition table and the risk gate; a
manifest that could set state would route straight around both. epictask
`dependsOn` edges are re-applied on every run, which is safe because the server
treats a duplicate edge as a no-op.
**Every feature ships smoke tests.** Adds the vertical: create an epic with
mixed outcome/quality criteria, round-trip it through get -o yaml → apply -f,
chain an epictask, then drive a task through the lifecycle asserting the risk
gate blocks `done` without a passing QA verdict, that an illegal transition is
refused, that history carries every transition and the verdict, and that reject
without a comment fails. Skips when mcpd is unreachable; a failure against a
reachable daemon means it predates the feature.
The round-trip assertion is the interesting one: it feeds `get`'s own output
straight back into `apply` and fails if the two surfaces have drifted, rather
than trusting that they agree.
Local: build clean, workspace 2457 passed, db 128 passed (same 4 pre-existing
Secret failures), lint back to the repo's baseline 794 errors after replacing
destructured discards with an explicit omit() helper.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01N4wNHWf7xSwnZCWpJcyv9p
A system agent, seeded like the system project and system skills, whose entire
job is running mcpctl for the user:
mcpctl chat mcpctl -m 'Make a prompt in the SRE project saying we run multiple
agents on the same source at once, so always use git worktrees'
## The permission model is the design
Every tool call loops back through mcpd's own HTTP API carrying the CALLING
USER's bearer token. The agent holds no credential of its own and never touches
the service layer directly, so the existing auth hook, RBAC hook, zod
validation, ownership scoping and audit trail all apply verbatim. It is exactly
as privileged as the person talking to it, never more.
Calling services directly would have been simpler and would have meant
re-implementing every one of those checks at a second call site — precisely the
RBAC carve-out this project refuses to make. A 403 is surfaced as a 403, not
retried or routed around; the system prompt tells the agent to report it.
Verified live, not just mocked: the same tool with an admin token created a
project, and with a plain user's token returned "Permission denied (403)".
No token at all refuses before any request is made; a bad token is rejected by
mcpd's own auth, because there is no ambient credential to fall back on.
## Shape
- `Agent.toolset` (TEXT, default 'project') selects the surface. 'mcpctl' means
no project prompts and no MCP tools — managing mcpctl is the whole job, and an
unrelated project's context would be noise it might act on.
- Five generic tools (list/get/create/update/delete + resource) rather than one
per resource. mcpctl is a REST API over named resources, so that is the honest
surface, and this repo's own tool-selection findings say a scoped catalogue
beats a large flat one.
- Names are resolved to ids by listing first, and a miss reports what IS
available, so the model never guesses an id.
- Resource is a closed enum: the model cannot invent an endpoint.
Also exercised end to end against a live mcpd: created the worktrees prompt in
the `sre` project, listed it back, and read its content — the user's original
example, working.
Tests: 17 unit tests covering the token plumbing, the no-token and 403 paths,
resource allowlisting, name resolution, and request shaping.
Local: build clean, workspace 2457 passed (+17), lint unchanged at 794 errors.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01N4wNHWf7xSwnZCWpJcyv9p
Everything so far was API-only. This adds the CLI surface and proves the whole
vertical works end to end against a real mcpd and a real Gitea.
mcpctl create gitprovider mysources --url https://mysources.co.uk \
--secret gitea-token --key TOKEN
mcpctl repos discover mysources --query mcp --project demo
mcpctl repos import mysources --project demo michal/mcpctl
mcpctl create epic ship-agents --project demo --content "..." \
--criterion "cli=CLI can drive a task" --criterion "q:feel=reads at a glance"
mcpctl create epictask foundation --epic ship-agents
mcpctl create task wire-cli --epictask foundation --title "Wire the CLI"
mcpctl tasks advance wire-cli in_progress
mcpctl tasks verdict wire-cli --stage qa_review --commit 3f2ab19 --pass --summary "..."
mcpctl tasks accept wire-cli --note "ship it"
mcpctl tasks history wire-cli
`repos` and `tasks` are their own verb groups rather than more create/edit
flags. Discovery is inherently two-step — look at what is there, then pick — and
accept/reject are the human gate, not field writes.
`get` gained columns for all five resources. The Epic row shows completeness as
"40% (2/5)" and CodeRepo shows SOURCE/FROM, so imported repos say where they
came from at a glance.
BREAKING: `mcpctl get tasks` now lists WORK ITEMS. The inference queue keeps
`inference-task(s)` and gains `itask(s)`; the short `task`/`tasks` aliases move
to work items. This completes the split already made in RBAC (a186d95) — having
the CLI and RBAC disagree about what "tasks" means would be worse than the
rename.
Verified end to end, not just unit tested: mcpd run locally against the test
database, admin bootstrapped, then the full flow driven through the built CLI.
Discovery returned the real repositories from https://mysources.co.uk, import
created the CodeRepo with provenance, re-discovery marked it `imported`, and the
task walked todo → in_progress → building → self_testing → qa_review with a
FAILING QA verdict correctly refusing `done`, then bounced back, passed on a
second commit, and merged through the accept gate. `tasks history` shows both
verdicts and all twelve transitions.
Two bugs the walkthrough caught that unit tests had not: `get repos` and
`get epics` crashed on `padEnd` because Prisma includes hand back `{ name }`
rather than a bare string, and the completions guard test failed because the
resource lists had changed without regenerating.
Local: build clean, workspace 2440 passed, db 128 passed (same 4 pre-existing
Secret failures), lint errors unchanged at the repo's pre-existing 794 (+2
no-console warnings, the same shape every other command registration uses).
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01N4wNHWf7xSwnZCWpJcyv9p
Two resources landing together, because a repo you have to hand-type is a repo
nobody adds.
CodeRepo is a git repository attached to a project. GitProvider is a configured
forge instance you register once — URL and credential in one place — and then
list and import from.
Nothing above the driver layer knows what a Gitea is. src/mcpd/src/services/git/
holds the GitProviderDriver interface, the Gitea implementation and a registry
that resolves by type, mirroring LlmAdapterRegistry.build. Adding GitHub or
GitLab is one file plus one case. The instance URL always comes from the
GitProvider row; there is no hardcoded host anywhere.
Follows the mcpservertemplates convention: a declarative resource applied via
apply -f, with instances keeping denormalized provenance that outlives the
source. CodeRepo stores sourceProvider + sourceFullName alongside the FK, so
"where did this repo come from" stays answerable after the forge is
disconnected — the same reasoning as McpServer.templateName/templateVersion.
The FK is SetNull, so disconnecting a forge never deletes the repos, or the
tasks hanging off them, that came from it.
Discovery uses /repos/search, deliberately NOT /user/repos. The latter needs the
read:user scope, which repo-scoped automation tokens do not carry — the token
in this repo's .env has write:repository and 403s on /user/repos. The driver
also passes Gitea's 403 scope message through verbatim, so an under-scoped
token says which scope it is missing instead of looking like an empty forge.
Import is partial-success by contract: one bad name must not lose the other
nineteen. Everything importable is imported; everything skipped comes back with
a reason, and a re-import is distinguished from a name collision with an
unrelated repo. Imported repos store no credential of their own and inherit the
forge's, so rotating one token covers every repo from that forge.
Credentials never leave as plaintext: reads return the reference only, and
resolveCheckouts() — the single path that produces a token — exists for the
session provisioner and is called by no route. References are verified at
definition time, so a wrong key fails with "secret has no key 'X'" rather than
as a clone that mysteriously cannot authenticate.
Clone URLs must be https/ssh/scp-style and checkout paths must be relative with
no ".." segment: both are inputs to the clone step, and neither should be able
to point it at mcpd's own filesystem or escape the workspace.
Tests: 18 CodeRepo + 22 GitProvider unit tests. Gitea driver additionally
verified live against https://mysources.co.uk (Gitea 1.25.4, 10 repos
discovered, query filter working).
Local: build clean, workspace 2440 passed (+40), db 128 passed (same 4
pre-existing Secret failures), lint unchanged at 869.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01N4wNHWf7xSwnZCWpJcyv9p
The work-item spine: repository, validation, service and routes for Task, plus
its verdicts and append-only transition log.
Transitions are compare-and-swap. The expected current state is part of the
WHERE clause and the affected-row count is the answer, so two agents racing to
advance the same task cannot both win — a read-then-write would let both
"succeed" and silently lose one. Losing the race is a 409 naming the state the
task is actually in, not a 500, because on a shared board that is a normal
outcome the caller retries. Same idiom as InferenceTaskRepository's claim.
QA / Security / SRE are stages inside the task's own loop, not post-submit
reviewers, so each can bounce straight back to in_progress — that is what
"feedback" means here. A failed build likewise goes backward, not forward.
`merged` allows only `reopened`, so a merged task cannot be quietly edited;
post-merge failure is the one way back in.
The gate on `done` scales with risk instead of running a fixed pipeline:
trivial needs nothing, normal needs a passing QA verdict, high needs QA +
Security + SRE. Forcing a one-line docs fix through security review is how a
process gets routed around, so the cost has to match the risk. A FAILING
verdict does not clear a gate — a recorded failure is evidence the gate was
tried and did not hold. The 400 names which gates are still missing.
The human accept gate is a task-level yes/no on a resolved outcome, not
per-tool approval: accept fires the merge and stores the comment, reject sends
it back to in_progress and REQUIRES a comment, because the comment is the
direction the agent works from.
GET /tasks/:id returns `legalTransitions` so no client has to hardcode the table.
Tests: 25 service tests — full happy path, every stage's bounce-back, illegal
and no-op transitions, the CAS race (including that no event is written when
the race is lost), the complete risk matrix including failing-verdict-does-not-
count, and the accept/reject gate.
Local: build clean, workspace 2400 passed (+25), db 128 passed (same 4
pre-existing Secret failures), lint unchanged at 869.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01N4wNHWf7xSwnZCWpJcyv9p
`mcpctl chat <agent>` failed with
HTTP 503 {"error":"service_unavailable","message":"Cannot reach mcpd daemon. Is it running?"}
while mcpd was answering /healthz in 32ms. The message was wrong in a way that
cost real debugging time: mcplocal was reaching mcpd fine and giving up after
30s. journalctl shows the signature plainly — statusCode 503 with
responseTime 30003.87 on POST /api/v1/agents/reviewer/chat.
This blocks the agentic-teams epic outright. An agent turn is a multi-turn
tool-use loop that runs for minutes by design, so a 30s ceiling on the chat path
is not a safety net, it is a guaranteed failure for every non-trivial turn.
Three defects, all in the same path:
1. One blanket budget for every forwarded route. DEFAULT_TIMEOUT_MS = 30_000 is
right for CRUD and wrong for chat. Chat, project chat, llm infer and
inference-task streams now get LONG_RUNNING_TIMEOUT_MS (600_000, override
with MCPLOCAL_LONG_TIMEOUT_MS) — matching STREAM_TIMEOUT_MS, which the CLI
already allowed. mcplocal in the middle was the binding constraint.
2. Timeouts were reported as connection failures. Split UpstreamTimeoutError
out of ConnectionError and map it to 504 with an accurate message that says
the daemon IS reachable. ConnectionError still means unreachable and still
returns 503. Verified nothing else branches on ConnectionError.
3. SSE was buffered. `forward()` reads the whole body through res.text(), so
even turns that finished in time arrived as one blob and the CLI's live
token output never appeared. Streaming routes now use forwardStream() and
pipe the body straight through, preserving content-type and
x-accel-buffering (dropping the latter lets intermediaries re-buffer and
reintroduces the stall).
Also closes the escape that produced the sibling `500 code:23` failure: the body
read in forward() was outside the try, so when mcpd had already written SSE
headers the raw DOMException reached Fastify unhandled.
Tests: 9 new proxy tests. The two that matter — "does not abort an agent chat
that outlives the CRUD budget" and "streams SSE through instead of buffering" —
were confirmed to FAIL against the pre-fix behaviour and pass after. Three
existing mcpd-client tests asserted the old taxonomy and were updated to assert
the new one deliberately.
Local: build clean, workspace 2375 passed, lint unchanged at 869.
NOT YET LIVE: mcplocal runs from the installed RPM, so this needs a package
rebuild + `systemctl --user restart mcplocal` to take effect.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01N4wNHWf7xSwnZCWpJcyv9p
Roadmap nodes plus the edges that chain them into the dependency graph the
design doc calls the spider web. Edge direction is "from depends on to".
The database cannot enforce graph integrity here — production provisions with
`prisma db push`, so a trigger or CHECK would never exist there, and the schema
test added earlier documents that boundary explicitly. So the service rejects
the three things that would otherwise corrupt the roadmap:
- self-edges,
- edges crossing Epics, since a roadmap is one Epic's graph,
- cycles, which make the graph unschedulable and hang any topological walk.
Cycle detection is a BFS over the prerequisite direction: adding from→to closes
a loop exactly when `to` already depends transitively on `from`. The error names
the offending path, not just "cycle", so a failed apply says which nodes to look
at. The traversal carries a seen-set that also guarantees termination if the
stored graph ALREADY contains a cycle — from a hand-edited database or an
earlier bug — rather than spinning forever. There is a test for that case.
Adding an existing edge is a no-op rather than an error, so applying a manifest
twice is safe.
GET /api/v1/epics/:id/roadmap returns nodes with per-node completeness, both
edge directions, and `ready`: the nodes whose prerequisites are all satisfied,
which is the question anyone actually asks of a roadmap. A cancelled
prerequisite counts as satisfied — it is not coming. Completeness reuses
EpicService.completeness verbatim so a node and its Epic can never disagree
about what "met" means.
Tests: 19 service tests — the full cycle matrix (self, 2-cycle, long transitive,
diamond-is-fine, pre-existing-cycle termination), roadmap edge directions,
readiness including the cancelled-prerequisite case, and CRUD guards.
Local: build clean, workspace 2366 passed (+19), db 128 passed (same 4
pre-existing Secret failures), lint unchanged at 869.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01N4wNHWf7xSwnZCWpJcyv9p
Full CRUD for Epic: repository, zod validation, service, routes, main.ts
wiring, and revision tracking (ResourceType gains 'epic', which is a plain
string union by design so this needed no migration).
The substance is the done rule, not the CRUD. An Epic exists to stop "every
task completed, still no car", so status is deliberately NOT settable through
PUT — it goes through POST /epics/:id/status, which enforces:
- `delivered` requires every done-criterion met. The 400 names the unmet ids
(truncated past 5) so the caller knows what is actually missing.
- `delivered` with recorded tradeoffs is refused — if you accepted a gap, the
status has to say so.
- `delivered_with_tradeoffs` requires at least one recorded tradeoff, otherwise
it is `delivered` wearing a disguise.
- `delivered` with zero criteria is refused: nothing was claimed, so nothing
can have been delivered.
- `cancelled` and `open` are always allowed — abandoning needs no ceremony, and
reopening is how a gap gets fixed.
Completeness counts `quality` criteria alongside `outcome` ones. Grading only
the mechanically checkable half is exactly the failure the Epic exists to
prevent, so both feed the percentage, and it is computed server-side and
attached to every response rather than left for clients to re-derive.
Tradeoffs are append-only and stamped with who and when — the record is the
audit trail for why "done" was honest.
Scope changes patch-bump semver, so creep is versioned as visibly as
under-delivery.
Tests: 26 service tests (the full done-rule matrix, tradeoff append/stamping,
criterion flipping, semver bumping, and that a plain update cannot smuggle a
status through) and 11 route tests (completeness attachment, owner stamping,
400 vs 404 vs 500 behaviour, the status gate over HTTP).
Local: build clean, workspace 2347 passed (+37), lint unchanged at 869.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01N4wNHWf7xSwnZCWpJcyv9p
An MCP server container is long-lived and restarts forever; an agent session pod
is the opposite — it does a unit of work and dies. ContainerSpec could not
express that: restartPolicy was hardcoded 'Always' in generatePodSpec, and there
were no volume, workdir, or init-container fields at all.
Adds four optional fields to ContainerSpec (restartPolicy, workdir, volumes,
initContainers) and implements them in both backends. All optional, so every
existing caller is byte-identical — generatePodSpec still emits 'Always' unless
asked otherwise, and omits volumes/initContainers keys entirely when unused.
Kubernetes gets this nearly free: initContainers run to completion in
declaration order before any app container starts, so "clone the repos, then
work them" needs no sequencing logic from us. generateDeploymentSpec
deliberately ignores spec.restartPolicy — k8s rejects anything but 'Always' on a
Deployment pod template — and says so in a comment.
Docker has no init containers, so they are emulated as sequential runs sharing
the same volumes. The important property is the failure path: a non-zero init
exit aborts BEFORE the main container is created, and the error carries the init
container's last 50 log lines. Otherwise a failed git clone would leave an empty
workspace that looks exactly like a successful one, and the agent would start
work against nothing. Named Docker volumes are namespaced by container name to
stand in for emptyDir.
Tests: 10 k8s manifest tests (defaults unchanged, each restart policy, volume
and mount emission, init ordering, name sanitization, security context still
hardened) and 9 Docker tests including the abort-before-main case and log
propagation.
Local: build clean, workspace 2310 passed (+19), lint unchanged at the repo's
pre-existing 869.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01N4wNHWf7xSwnZCWpJcyv9p
Adds the foundation entities from docs/agentic-teams.md §5/§7: a versioned
vision layer (Epic -> EpicTask -> Task), git repositories attached to projects,
agents as many-to-many project members, and the ephemeral session that IS an
agent's workspace.
9 tables + 5 enums, purely additive — no drops and no column changes on
existing tables, since every new relation is held on the new side.
Shape notes:
- EpicTask chaining uses an explicit EpicTaskEdge table rather than a
self-referencing array, so the roadmap graph walks in both directions and
edges can carry a type later.
- TaskVerdict is commit-pinned and immutable: "QA verified 3f2ab" stays
reproducible, which is what lets later coverage claims be audited rather than
believed. TaskEvent is the append-only transition log.
- Task survives losing its repo and its assignee (both SetNull) — a work item
outlives whoever happened to be working it — but cascades from its EpicTask.
- CodeRepo, not Repository: `repository.repository.ts` next to 24 other
*.repository.ts data-access files would be actively confusing. CLI resource is
`repos`, with repo/repository/repositories aliased onto it.
- AgentProject.llmId exists so the brain-ownership question
(docs/agentic-teams.md §15 Q1) can be settled either way without a migration.
- Agent.projectId is retained and coexists with AgentProject for one release so
findByProjectId / prepareContext keep working. Note this partially reverses
20260223120000_remove_project_members — deliberately, and only for agents;
human membership stays RBAC-driven.
No CHECK constraints anywhere, on purpose: deploy/entrypoint.sh provisions with
`prisma db push`, so hand-written migration SQL never runs in production. The
migration file is kept for local `migrate dev` parity and as a readable record,
and says so in its header. Invariants that cannot be expressed in the schema
(EpicTaskEdge cycle rejection) belong to the service layer; a schema test
documents that boundary rather than pretending the database enforces it.
Tests: 23 schema tests covering defaults, per-project name scoping, cascade vs
SetNull on every FK, all 12 lifecycle states, and the GC-sweep predicate.
clearAllTables extended for all 9 models.
Verified locally against postgres:16-alpine on :5433 — db suite 128 passed
(baseline was 105; the same 4 pre-existing Secret_backendId_fkey failures
remain, unrelated). Workspace 2291 passed, build clean, lint unchanged at the
repo's pre-existing 869.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01N4wNHWf7xSwnZCWpJcyv9p
v5 gave the durable inference queue the RBAC resource `tasks`, on the
reasoning that inference was the only kind of task we had. Agentic teams
introduces work-item tasks (Epic -> epictask -> task), which need that name.
The queue moves to its own `inference-tasks` resource, matching its existing
/api/v1/inference-tasks route prefix, and `task`/`tasks` now mean work items.
Also lands the agentic-teams aliases: epic, epictask/et, repo/repository, session.
The rename is not purely a code change. ResourceBindingSchema runs
normalizeResource at parse time and stores the normalized string, so every
RbacDefinition row written before the split literally contains "tasks" meaning
inference. Shipping the alias change alone would silently stop those grants
authorizing the queue and start them authorizing work items -- a privilege
change in both directions. bootstrap/rbac-tasks-backfill.ts rewrites them once
at startup, alongside migrateAdminRole.
The run-once marker is claimed BEFORE the rewrite, not after. A failed marker
write after a successful rewrite would let the next startup rewrite again,
converting genuine post-split work-item grants into inference grants. Losing
the rewrite is recoverable by re-granting; running it twice silently is not --
so if the marker cannot be persisted we skip the rewrite and warn loudly.
mapUrlToPermission moves in lockstep: /api/v1/inference-tasks maps to
`inference-tasks`, otherwise the split would break queue authorization on the
first request.
Adds generic getBooleanSetting/setSetting helpers to bootstrap/system-settings.ts,
following the existing "global settings live in a Secret" pattern. setSetting
refuses to write when the settings secret has been migrated onto a non-plaintext
backend, where a direct `data` write would read back empty.
Out of scope, tracked separately: the CLI still routes `mcpctl get tasks` to the
inference queue (src/cli/src/commands/shared.ts). Repointing it needs the
work-item resource to exist first, or the command breaks with no replacement.
Tests: 14 unit tests cover alias resolution, the rewrite, wildcard and
operation bindings being left alone, idempotency, and both marker-failure paths.
Full workspace suite green (2291 tests). The new smoke test is RED until mcpd
is deployed with this change -- it asserts post-split storage against a live
daemon, and the currently deployed one predates the split.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01N4wNHWf7xSwnZCWpJcyv9p
Converts docs/agentic-teams.md §16 steps 1-3 into a trackable PRD and 48
appended tasks (IDs 115-162) in the master tag. Steps 4-8 land as 5 low
priority placeholders wired onto the spine, pending the §15 open questions.
The PRD is grounded in the current codebase rather than the design doc's
indicative sketches, which changed the shape of the work:
- RBAC `tasks` is already owned by the inference queue, and bindings are
normalized at parse time, so existing RbacDefinition rows literally hold
"tasks". Taking the name for work items needs a one-shot guarded backfill
to `inference-tasks` (task 116).
- deploy/entrypoint.sh uses `prisma db push`, so migration SQL never runs in
production. Invariants are service-layer, not CHECK constraints.
- Ephemeral pods do not exist: restartPolicy is hardcoded Always and
ContainerSpec has no volume/workdir fields (tasks 140-142).
- `exec` is cheap: KubernetesOrchestrator already implements exec/attach over
k8s.Exec/k8s.Attach; only an API + CLI surface is missing (tasks 144-145).
- No git provider client exists in product code (task 148).
- ChatService caps its tool loop at 12 iterations, so the AgentRuntime runs a
headless coding agent in-pod rather than extending chat (tasks 146-147).
Names the git-repo entity CodeRepo (CLI `repos`) to avoid colliding with the
repository/ data-access layer; AgentProject carries an optional llmId so the
brain-ownership question can be settled either way without a migration.
Generated task_*.md files are left untracked, matching how this repo has
treated them since the first commit.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01N4wNHWf7xSwnZCWpJcyv9p
One slow call stalled every later request because the stdin loop awaited each in turn; queued requests were never sent so nothing could time them out and the client saw silence. Dispatch is now concurrent. Regression test verified to fail on the old code.
The bridge's stdin loop awaited every request before reading the next line, so
it handled exactly one at a time. A single slow call therefore stalled every
later request — and because those requests were never even sent, nothing could
time them out. The client saw silence, not an error.
Observed 2026-08-05: two gitea calls through this bridge sat completely mute
until Claude Code aborted them at its own 1800s idle limit, reporting "sent no
response or progress". The upstream was healthy the whole time — a fresh
session answered the same tools in 0.3s, and the gitea MCP server's own log
showed the calls never reached it. They died queued in the bridge.
JSON-RPC ids exist precisely so responses may return out of order, so nothing
here needed a queue. Requests now dispatch concurrently and are tracked in a
set; stdin close awaits them before the session DELETE, otherwise a concurrent
call races the teardown and dies with a 404.
We still serialise until the session id exists: it arrives on the first
response, and firing later requests without it would open a second upstream
session. A client sends `initialize` first and waits for the reply anyway, so
this costs one round trip rather than throughput.
Also makes the per-request timeout configurable via MCPCTL_MCP_TIMEOUT_MS
(default unchanged at 30s) and names it in the timeout error, so a project with
genuinely long tool calls can raise it instead of hitting a hardcoded wall.
Tests pin both halves: a fast request must overtake a slow one (this fails on
the old serial code — verified by reverting), and a failed request must still
produce a JSON-RPC error carrying the original id rather than nothing.
Even "<=8 words" was ignored — the model wrote a full-sentence reasoning, pushing
begin_session to ~7s (thin under the 8s budget). The gate only needs the names, so
request just {"selectedNames":[...]}. Validated live vs vllm-current
(glm-4.6-reap-fast): 1.5s, valid JSON. (reasoning defaults to '' in extractSelection.)
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
With the fast (no-think) route, gate-selection latency is now proportional to the
ANSWER length (a no-think model runs ~10-13 tok/s), not hidden thinking. A verbose
"reasoning" field pushed a selection to ~9s (over the 8s budget). Request compact
JSON with a <=8-word reasoning → ~2s (vs ~9s prose, or ~1.2s with no reasoning).
Validated live against vllm-fast (glm-4.6-reap-fast): 2.0s, valid selection JSON,
zero reasoning tokens.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Lets ops route gate prompt-selection at a fast (no-think) server Llm independent
of the project's chat llmProvider — so selection stays ~1-2s while chat keeps its
thinking model. Overrides the project llmProvider for the gate's server-selection
path; unset → prior behavior (use the project's llmProvider). Pairs with the
litellm qwen3-fast no-think alias (kubernetes-deployment) + a `vllm-fast` mcpd Llm.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Prompt selection is mechanical classification — it gains nothing from chain-of-
thought, and a reasoning model (qwen3-thinking) otherwise burns its whole budget
reasoning (~40s, 6k+ chars) and blows the gate's 8s timeout. The gate's server
selection request now includes thinking-suppression hints, forwarded verbatim by
mcpd's passthrough adapter to litellm/vLLM:
chat_template_kwargs.enable_thinking=false (Qwen3 hard-off)
reasoning_effort=low (o-series / newer vLLM)
Harmless for models that ignore them; if a backend rejects them the selector
falls back to the local provider. Override via MCPCTL_GATE_SELECT_EXTRA_BODY
(JSON), or '' to disable.
NOT yet validated live — the qwen3-thinking vLLM is crashlooping again (0/1,
HTTP 500), and whether litellm forwards chat_template_kwargs to vLLM is unconfirmed
(the /no_think prompt directive was NOT honored). Validate when the backend
recovers. mcpld gate/selector tests green (75).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Cloud/server keys belong at the k8s/mcpd level; mcplocal handles only the user's
personal tokens. The gate's prompt-selection used mcplocal's LOCAL provider
registry (heavy=anthropic = the user's OAuth token, which is API-gated and 404s),
so selection always degraded.
Now LlmPromptSelector tries sources in priority order: (1) the project's server
Llm via mcpd's inference proxy (POST /api/v1/llms/:name/infer) — cloud/server
keys stay at k8s — then (2) the local personal-token provider as fallback. First
source that returns valid selection JSON wins. Extracted pickCompletionText
(content ?? reasoning_content) + extractSelection helpers.
Wiring: GatePluginConfig.llmProvider (from the project), threaded via
createDefaultPlugin; handleBeginSession builds serverInfer from ctx.postToMcpd.
Verified live: POST .../llms/vllm-current/infer returns valid selection JSON
(qwen3-thinking, 4000-token budget). + selector tests (server preferred, local
fallback on error, server-only). mcplocal suite green (755).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The local mcplocal gate resolves heavy=[anthropic]; its OAuth-gated key 404s the
configured opus-4 model, and anthropic.ts (like openai.ts before this series)
JSON.parsed the error body as a completion → empty content → the gate's
misleading "did not contain valid selection JSON". Reject on status >= 400 with
"Anthropic HTTP <status>: <body>" so the true reason (model-not-found / auth)
surfaces as the degradedReason. + parity test.
Note: this makes the error honest; it does not make the local gate SUCCEED — that
needs a working heavy provider (the anthropic OAuth token is fundamentally gated).
The qwen3-thinking gate path (k8s mcplocal) is addressed by the budget+reasoning
fix in the previous commit.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The presented→canonical resolver was only built during tools/list, so a
favourite/<tool> call arriving before any list this session fell through to the
namespaced router and 404'd (all/ was fine — it prefix-strips). Real MCP clients
list first, but the smoke test's initialize→call flow exposed it. Extract
buildPresentation() and lazily rebuild the resolver in onToolCallBefore when a
favourite//all/ name isn't resolved yet. + unit test (favourite/ call with no
prior list) and realistic listTools() in the smoke routing case.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Two independent, live-reproduced causes of "Smart prompt-selection unavailable
(LLM response did not contain valid selection JSON)" in the gate:
- Cause B (openai.ts request): the response handler JSON.parsed the body with no
res.statusCode check, so a LiteLLM/vLLM HTTP 500 ("Connection error") parsed as
an empty completion and surfaced as generic "bad JSON", masking backend-down.
Now reject on status >= 400 with "OpenAI HTTP <status>: <body>" so the selector
reports the true degradedReason.
- Cause A (parseResponse + selector budget): a reasoning model (qwen3-thinking)
spends its whole 1024-token budget in reasoning_content and emits content=null,
so the {selectedNames} JSON never appears. parseResponse now falls back to
reasoning_content (incl. provider_specific_fields) when content is empty, and
the selector budget is raised 1024 -> 4000 (env MCPCTL_GATE_SELECT_MAX_TOKENS)
so thinking + the JSON answer both fit.
Note: mcpd's chat path got a reasoning_content fallback in PR #67; this is the
separate mcplocal gate-selection provider, which never did. Tests: openai-provider
(500 rejects, reasoning_content capture/preference). mcplocal suite green (749).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Measured winner from the DGX-Spark bake-off (toolsim.py, 145-tool catalog): a
curated favourite/<tool> shortlist + the full all/<server>/<tool> catalog + a
load-bearing "prefer favourite/ first" instruction nearly halved wander (37→20)
and 2.5x'd first-pick (2→5/8) vs a flat catalog. The instruction is load-bearing;
enriching descriptions did not help.
- New mcplocal plugin `favourite-index.ts`: composes AFTER gate (no-ops while
gated), reshapes the ungated upstream catalog into favourite/ + all/, injects
the instruction (onInitialize), and rewrites presented names back to canonical
server/tool in onToolCallBefore so normal routing + content-pipeline still run.
Gate/agent virtual tools pass through untouched; favourites are upstream-only.
- compose.ts: onInitialize now concatenates plugin instructions (was first-non-null)
so favindex can contribute its banner alongside the gate's.
- Per-project config `Project.favouriteIndex` {enabled, tools[], maxFavourites};
surfaced to the proxy via discovery; wired at project-mcp-endpoint when enabled.
- Usage derivation: mcpd tool-usage ranking over tool_call_trace events
(normalizing presented names → canonical), GET /api/v1/audit/tool-usage, and
`mcpctl favourites suggest|list`.
- CLI: `create project` gains --favourite/--favourite-index/--max-favourites;
favouriteIndex round-trips through get -o yaml | apply -f. Completions regenerated.
- Tests: plugin unit (presentation, rewrite routing, gated no-op, collisions),
compose merge, canonicalizeToolName, buildFavouriteIndex, + a live smoke test.
- Docs: docs/tool-presentation.md.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Record the empirical finding: a Claude subscription OAuth token is gated by
Anthropic (404/429 as a raw API) and can't serve as the server-side fallback,
even though mcpd's adapter now sends it via Bearer. The anthropic-fallback is
wired-but-inert until mcpd Secret anthropic-key holds a real console API key.
Also note the chain is now Pulumi-owned so pulumi up won't wipe it.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude subscription / Claude Code keys are OAuth tokens and authenticate with
Authorization: Bearer, not x-api-key. mcpd's server adapter only sent x-api-key,
so registering an OAuth key as a fallback Llm 401'd ("invalid x-api-key").
Mirror the mcplocal client adapter's isOAuth switch so an OAuth key works as a
server-side anthropic fallback.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
