feat(mcpd): inference proxy — POST /api/v1/llms/:name/infer

Why: the point of the Llm resource (Phase 1) is that credentials never leave
the server. This lands the proxy: clients POST OpenAI chat/completions to
mcpd, mcpd attaches the provider API key server-side, and the response
streams back as OpenAI-format SSE.

Design:
- Wire format client-side is always OpenAI chat/completions — every existing
  SDK speaks it. Adapters translate on the provider side.
- `openai | vllm | deepseek | ollama` → pure passthrough (they already speak
  OpenAI). `anthropic` → translator to/from Anthropic Messages API
  (system-string extraction, content-block flattening, SSE event remap).
- Plain fetch; no @anthropic-ai/sdk dep. Consistent with the OpenBao driver
  shape and keeps the proxy layer thin.
- `gemini-cli` intentionally rejected — subprocess providers need extra
  lifecycle plumbing; deferred to a follow-up.
- Streaming: adapters yield `StreamingChunk`s; the route frames them as
  `data: <json>\n\n` + terminal `data: [DONE]\n\n` so any OpenAI client
  works unchanged.

RBAC:
- New URL special-case in mapUrlToPermission: `POST /api/v1/llms/:name/infer`
  → `run:llms:<name>` (not the default create:llms). Users need an explicit
  `{role: 'run', resource: 'llms', [name: X]}` binding to call infer.
- Possession of `edit:llms` does NOT imply `run` — keeps catalogue
  management separate from spend.

Audit: route emits an `llm_inference_call` event per request (llm name,
model, user/tokenSha, streaming, duration, status). main.ts wires it to the
structured logger for now; hook is in place for a richer audit sink later.

Tests:
- 11 adapter tests (passthrough POST shape + default URLs + no-auth ollama +
  SSE forwarding; anthropic translate request/response + non-2xx wrap + SSE
  event translation; registry dispatch + caching + unsupported-provider).
- 7 route tests (404, 400, non-streaming dispatch + audit, apiKey failure,
  null apiKeyRef path, streaming SSE output, 502 on adapter error).
- Full suite 1830/1830 (+18 from Phase 1's 1812). TypeScript clean.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

src/mcpd/tests/llm-infer-route.test.ts

@@ -0,0 +1,208 @@
+import { describe, it, expect, vi, afterEach } from 'vitest';
+import Fastify from 'fastify';
+import type { FastifyInstance } from 'fastify';
+import { registerLlmInferRoutes } from '../src/routes/llm-infer.js';
+import { LlmAdapterRegistry } from '../src/services/llm/dispatcher.js';
+import { errorHandler } from '../src/middleware/error-handler.js';
+import type { LlmView } from '../src/services/llm.service.js';
+import { NotFoundError } from '../src/services/mcp-server.service.js';
+
+let app: FastifyInstance;
+
+function makeLlmView(overrides: Partial<LlmView> = {}): LlmView {
+  return {
+    id: 'llm-1',
+    name: 'claude',
+    type: 'anthropic',
+    model: 'claude-3-5-sonnet-20241022',
+    url: '',
+    tier: 'heavy',
+    description: '',
+    apiKeyRef: { name: 'anthropic-key', key: 'token' },
+    extraConfig: {},
+    version: 1,
+    createdAt: new Date(),
+    updatedAt: new Date(),
+    ...overrides,
+  };
+}
+
+afterEach(async () => {
+  if (app) await app.close();
+});
+
+function sseResponse(events: string[]): Response {
+  const body = events.join('\n\n') + '\n\n';
+  const stream = new ReadableStream<Uint8Array>({
+    start(controller) {
+      controller.enqueue(new TextEncoder().encode(body));
+      controller.close();
+    },
+  });
+  return new Response(stream, { status: 200 });
+}
+
+interface LlmServiceLike {
+  getByName: (name: string) => Promise<LlmView>;
+  resolveApiKey: (name: string) => Promise<string>;
+}
+
+async function setupApp(
+  llmService: LlmServiceLike,
+  adapters: LlmAdapterRegistry,
+  onInferenceEvent?: Parameters<typeof registerLlmInferRoutes>[1]['onInferenceEvent'],
+): Promise<FastifyInstance> {
+  app = Fastify({ logger: false });
+  app.setErrorHandler(errorHandler);
+  const deps: Parameters<typeof registerLlmInferRoutes>[1] = {
+    // eslint-disable-next-line @typescript-eslint/no-explicit-any
+    llmService: llmService as any,
+    adapters,
+  };
+  if (onInferenceEvent !== undefined) deps.onInferenceEvent = onInferenceEvent;
+  registerLlmInferRoutes(app, deps);
+  await app.ready();
+  return app;
+}
+
+describe('POST /api/v1/llms/:name/infer', () => {
+  it('returns 404 when the Llm does not exist', async () => {
+    const svc: LlmServiceLike = {
+      getByName: async () => { throw new NotFoundError('Llm not found: missing'); },
+      resolveApiKey: async () => '',
+    };
+    await setupApp(svc, new LlmAdapterRegistry());
+    const res = await app.inject({
+      method: 'POST',
+      url: '/api/v1/llms/missing/infer',
+      payload: { messages: [{ role: 'user', content: 'hi' }] },
+    });
+    expect(res.statusCode).toBe(404);
+  });
+
+  it('returns 400 when messages is missing', async () => {
+    const svc: LlmServiceLike = {
+      getByName: async () => makeLlmView({ apiKeyRef: null }),
+      resolveApiKey: async () => '',
+    };
+    await setupApp(svc, new LlmAdapterRegistry());
+    const res = await app.inject({
+      method: 'POST',
+      url: '/api/v1/llms/claude/infer',
+      payload: {},
+    });
+    expect(res.statusCode).toBe(400);
+  });
+
+  it('dispatches non-streaming to the adapter and returns its JSON', async () => {
+    const fetchFn = vi.fn(async () => new Response(JSON.stringify({
+      id: 'msg_1', model: 'claude-3-5-sonnet-20241022', role: 'assistant',
+      content: [{ type: 'text', text: 'hello' }],
+      stop_reason: 'end_turn',
+      usage: { input_tokens: 1, output_tokens: 1 },
+    }), { status: 200 }));
+    const adapters = new LlmAdapterRegistry({ fetch: fetchFn as unknown as typeof fetch });
+    const svc: LlmServiceLike = {
+      getByName: async () => makeLlmView(),
+      resolveApiKey: async () => 'sk-ant-xyz',
+    };
+    const events: unknown[] = [];
+    await setupApp(svc, adapters, (e) => events.push(e));
+
+    const res = await app.inject({
+      method: 'POST',
+      url: '/api/v1/llms/claude/infer',
+      payload: { messages: [{ role: 'user', content: 'hi' }] },
+    });
+    expect(res.statusCode).toBe(200);
+    const body = res.json<{ choices: Array<{ message: { content: string } }> }>();
+    expect(body.choices[0]!.message.content).toBe('hello');
+
+    // Audit event emitted
+    expect(events).toHaveLength(1);
+    expect((events[0] as { kind: string; llmName: string; status: number }).kind).toBe('llm_inference_call');
+    expect((events[0] as { llmName: string }).llmName).toBe('claude');
+    expect((events[0] as { streaming: boolean }).streaming).toBe(false);
+    expect((events[0] as { status: number }).status).toBe(200);
+  });
+
+  it('500s when apiKey resolution fails', async () => {
+    const adapters = new LlmAdapterRegistry();
+    const svc: LlmServiceLike = {
+      getByName: async () => makeLlmView(),
+      resolveApiKey: async () => { throw new Error('secret not found'); },
+    };
+    await setupApp(svc, adapters);
+
+    const res = await app.inject({
+      method: 'POST',
+      url: '/api/v1/llms/claude/infer',
+      payload: { messages: [{ role: 'user', content: 'hi' }] },
+    });
+    expect(res.statusCode).toBe(500);
+    expect(res.json<{ error: string }>().error).toMatch(/secret not found/);
+  });
+
+  it('skips apiKey resolution when the Llm has no apiKeyRef', async () => {
+    const fetchFn = vi.fn(async () => new Response(JSON.stringify({ id: 'x', choices: [] }), { status: 200 }));
+    const adapters = new LlmAdapterRegistry({ fetch: fetchFn as unknown as typeof fetch });
+    const resolveSpy = vi.fn();
+    const svc: LlmServiceLike = {
+      getByName: async () => makeLlmView({ type: 'ollama', url: 'http://ollama:11434', apiKeyRef: null }),
+      resolveApiKey: resolveSpy as unknown as LlmServiceLike['resolveApiKey'],
+    };
+    await setupApp(svc, adapters);
+
+    const res = await app.inject({
+      method: 'POST',
+      url: '/api/v1/llms/ollama-local/infer',
+      payload: { messages: [{ role: 'user', content: 'hi' }] },
+    });
+    expect(res.statusCode).toBe(200);
+    expect(resolveSpy).not.toHaveBeenCalled();
+  });
+
+  it('streams SSE chunks for stream: true', async () => {
+    const fetchFn = vi.fn(async () => sseResponse([
+      'event: content_block_delta\ndata: {"type":"content_block_delta","delta":{"type":"text_delta","text":"hi"}}',
+      'event: message_stop\ndata: {"type":"message_stop"}',
+    ]));
+    const adapters = new LlmAdapterRegistry({ fetch: fetchFn as unknown as typeof fetch });
+    const svc: LlmServiceLike = {
+      getByName: async () => makeLlmView(),
+      resolveApiKey: async () => 'sk-ant-xyz',
+    };
+    const events: Array<{ streaming: boolean; status: number }> = [];
+    // eslint-disable-next-line @typescript-eslint/no-explicit-any
+    await setupApp(svc, adapters, ((e: any) => events.push(e)) as any);
+
+    const res = await app.inject({
+      method: 'POST',
+      url: '/api/v1/llms/claude/infer',
+      payload: { messages: [{ role: 'user', content: 'hi' }], stream: true },
+    });
+    expect(res.statusCode).toBe(200);
+    expect(res.body).toContain('data:');
+    expect(res.body).toContain('[DONE]');
+    expect(events).toHaveLength(1);
+    expect(events[0]!.streaming).toBe(true);
+  });
+
+  it('502s on adapter errors (non-streaming)', async () => {
+    const fetchFn = vi.fn(async () => { throw new Error('upstream down'); });
+    const adapters = new LlmAdapterRegistry({ fetch: fetchFn as unknown as typeof fetch });
+    const svc: LlmServiceLike = {
+      getByName: async () => makeLlmView({ type: 'openai', url: 'http://example', apiKeyRef: null }),
+      resolveApiKey: async () => '',
+    };
+    await setupApp(svc, adapters);
+
+    const res = await app.inject({
+      method: 'POST',
+      url: '/api/v1/llms/x/infer',
+      payload: { messages: [{ role: 'user', content: 'hi' }] },
+    });
+    expect(res.statusCode).toBe(502);
+    expect(res.json<{ error: string }>().error).toMatch(/upstream down/);
+  });
+});