feat(pulumi): @mcpctl/pulumi — generic Pulumi provider for mcpctl
Some checks failed
CI/CD / lint (pull_request) Successful in 1m3s
CI/CD / typecheck (pull_request) Successful in 2m5s
CI/CD / test (pull_request) Successful in 1m19s
CI/CD / smoke (pull_request) Failing after 2m46s
CI/CD / build (pull_request) Successful in 2m25s
CI/CD / publish (pull_request) Has been skipped

New workspace package: a Pulumi dynamic provider (TypeScript, in-process) that
manages mcpctl resources declaratively from any Pulumi program. Motivated by
model drift: when the deployed vLLM/LiteLLM model changes, mcpctl's llm target
must follow, and a Pulumi resource makes that automatic on `pulumi up`.

- Generic core `McpctlResource({ kind, name, spec, mcpd })` round-trips ANY
  mcpctl resource kind; typed `Llm` wrapper for ergonomics.
- Engine talks direct HTTP to mcpd's REST API (ported from cli/api-client.ts),
  mirroring apply's name-keyed upsert (PUT strips immutable name/type) and
  get -o json's read field-stripping. No mcpctl binary needed → CI-friendly.
- CRUD/diff/check with correct replace semantics (name/type/kind/mcpd.url force
  delete-before-replace), 429 retry, secret token in state.
- Validation deferred to mcpd's Zod schemas, so new resource kinds work with no
  provider release.
- CommonJS build so the serialized dynamic provider's module refs are captured.
- 18 Vitest tests (mocked mcpd + Pulumi mocks); build + lint clean; README with
  auth (PAT) setup and consumer example for kubernetes-deployment.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This commit is contained in:
Michal
2026-07-17 22:02:22 +01:00
parent e2db8d9003
commit f3d92235db
16 changed files with 2222 additions and 10 deletions

96
src/pulumi/README.md Normal file
View File

@@ -0,0 +1,96 @@
# @mcpctl/pulumi
A [Pulumi](https://www.pulumi.com/) dynamic provider for [mcpctl](../../). Manage
mcpctl resources — `llm`, `server`, `project`, `secret`, `agent`, and any other
kind mcpctl supports — declaratively from any TypeScript Pulumi program.
It talks **directly to mcpd's REST API** (the same endpoints the `mcpctl` CLI
uses) and mirrors the CLI's two round-trip guarantees:
- `apply`'s name-keyed **upsert** (look up by `name`; PUT with immutable fields
stripped if it exists, else POST), and
- `get -o json`'s field-stripping on read, so state does not churn.
No `mcpctl` binary or `~/.mcpctl` credentials are required on the machine
running `pulumi up`.
## Install
```bash
npm install @mcpctl/pulumi @pulumi/pulumi
```
## Usage
### Typed `Llm` wrapper
```ts
import * as pulumi from "@pulumi/pulumi";
import * as mcpctl from "@mcpctl/pulumi";
const cfg = new pulumi.Config();
const mcpd = { url: cfg.require("mcpdUrl"), token: cfg.requireSecret("mcpdPat") };
new mcpctl.Llm("homelab-qwen", {
name: "qwen3-thinking", // upsert key (immutable — changing it replaces)
type: "openai", // immutable
model: servedModelName, // pulumi.Input<string> — e.g. from another resource
url: "http://litellm.nvidia-nim.svc.cluster.local:4000",
tier: "heavy",
mcpd,
});
```
When `model` (or `url`, `tier`, `description`, ...) changes, Pulumi issues an
in-place `PUT /api/v1/llms/qwen3-thinking` on the next `up`. Changing `name` or
`type` forces a replacement (delete-before-replace, since `name` is the identity).
### Generic resource — any kind
```ts
new mcpctl.McpctlResource("my-server", {
kind: "server",
name: "my-server",
spec: { /* sent verbatim to mcpd; mcpd's Zod schema validates it */ },
mcpd,
});
```
The provider does no per-kind schema validation beyond the name shape and the
required connection — it defers to mcpd. New mcpctl resource kinds work the day
mcpd ships them, without a provider release.
## Auth
The provider authenticates with a **long-lived PAT** (`mcpctl_pat_…`), passed as
`mcpd.token` (use `cfg.requireSecret(...)`).
Mint one with resource-wide `view:llms` + `edit:llms` bindings, e.g. via
`mcpctl apply -f -`:
```yaml
kind: mcptoken
name: pulumi-automation
projectId: <any-project-id> # currently required (see note)
expiresAt: null # never expires
bindings:
- { role: view, resource: llms }
- { role: edit, resource: llms }
```
The bindings are resource-wide, so the token drives `/api/v1/llms/*` regardless
of the project it is attached to (the project-scope guard only gates
`/api/v1/projects/*`). Grant broader bindings if you manage other kinds.
> **Note:** mcpd currently requires `projectId` when minting an mcptoken, so a
> management-scope PAT must attach to some (throwaway) project. A future mcpd
> change may make `projectId` optional for management tokens.
## Notes
- **Secrets in state:** dynamic-provider inputs (including the token) are stored,
encrypted, in Pulumi state. Use an encrypted state backend.
- **Serialization:** this package is CommonJS on purpose — Pulumi serializes the
provider object, and the CJS build lets its module references (`./engine`,
`./kinds`) be captured and re-required at runtime. Validate end-to-end with a
real `pulumi up` against a disposable mcpd before relying on it in production.