Conversation
|
🚅 Deployed to the rivet-pr-4396 environment in rivet-frontend
|
PR Review: Fix mcp-hub to serve built docs metadataOverall this is a clean, well-motivated fix. The core change — replacing the fake static corpus and the brittle workspace-import fallback with the real Astro-built output — is the right direction. A few things to flag: Indentation DriftSeveral files have inconsistent indentation that looks like a merge/rebase artifact:
const engine = createSearchEngine(metadata); // extra indent
const results = engine.search("", { ... }); // extra indent
expect(results.results).toHaveLength(0); // less indentThis looks like a rebase conflict artifact.
These are cosmetic but they will likely trip up auto-formatters and make future diffs noisy. Docker Production Stage:
|
Description
Replace the mcp-hub sample corpus path with the real website metadata build output, simplify the Docker image to build the website directly, and remove the fake semantic/hybrid search modes from
docs.search. This also updates metadata heading slug generation to match the website so the MCP server can load the real docs corpus without duplicate resource IDs.Type of change
How Has This Been Tested?
Ran
pnpm install --no-frozen-lockfile,SKIP_TYPECHECK_CODE_BLOCKS=1 pnpm --filter rivet-website run build,pnpm --filter @rivetkit/mcp-hub test,pnpm --filter @rivetkit/mcp-hub run build, a local MCP initialize/search smoke test against the built docs corpus, anddocker build -f rivetkit-typescript/packages/mcp-hub/Dockerfile ..Checklist: