2 votes · 2 up · 0 down
Backend repo is live: https://github.com/skridlevsky/openchaos-backend
TypeScript + pnpm, structured logging, env validation, CI pipeline. Health check at /health. Ready for PRs.
Port the MCP server over when you're ready. The in-memory cache will actually work there — persistent process, no cold starts.
YES.
Yes yes yes yes yes. Yup. Uh-huh. Affirmative. Indeed. Absolutely. Certainly. Positively. Without a doubt. You bet. Roger that. 10-4. Copy that. Understood. Acknowledged. Confirmed. Agreed. Correctamundo. Yessiree. Aye aye. Da. Oui. Sí. Ja. Hai.
👍 💯 ✅
All technical points addressed. One clean commit. Build passing. Standing by for openchaos-backend setup.
Will port everything over with proper SDK integration when the VPS is ready. Just ping me (and our robot overlords) when it's time to migrate to the promised land of persistent processes and actual cache hits.
Until then, this PR shall rest. 🤖
Cool concept. MCP access for AI agents fits OpenChaos perfectly. The tools are well-structured and the error handling is thorough.
This would work better as a standalone service. Here's the thinking:
I'm standing up a VPS backend for OpenChaos (see #152 for context). Your in-memory cache actually works on a persistent process, but on Vercel serverless it gets a near-zero hit rate because each request can be a cold start. The VPS gives you real caching and a long-running process, which is exactly what an MCP server needs.
A few technical notes for the potential restructure:
Unused SDK. @modelcontextprotocol/sdk is in package.json but never imported in any source file. The route handler manually implements JSON-RPC dispatching. On the VPS you could use the SDK's built-in StreamableHTTPServerTransport for proper Streamable HTTP support.
getOpenPRs() changes. The added merge status + commit status calls add 2 extra API requests per PR. With 20 open PRs that's 40 additional calls per page load. Worth splitting into a separate enriched function so the frontend isn't affected.
Dead code. predict-next-merge.ts is built but never exported or registered. Either wire it up or cut it.
Duplicated code. getHeaders() and GITHUB_REPO are defined in multiple tool files but already exist in github.ts. Worth deduplicating.
Path forward: Would you be interested in PR-ing the MCP tools into https://github.com/skridlevsky/openchaos-backend?
The tools themselves are solid and can move over mostly as-is. I'm also considering adding llms.txt to the main site so AI agents can discover the MCP endpoint.
🤖 OpenChaos Bot
Summary: This PR adds the @modelcontextprotocol/sdk and zod packages to the project, implying an integration with the Model Context Protocol and data validation using Zod. These libraries add significant backend functionality.
Files changed: 1 (package-lock.json)
Vibe: They're adding what now? This reads like someone threw a pile of dependencies at the wall and hoped our democracy would stick them together.
Backend repo is live: https://github.com/skridlevsky/openchaos-backend
TypeScript + pnpm, structured logging, env validation, CI pipeline. Health check at /health. Ready for PRs.
Port the MCP server over when you're ready. The in-memory cache will actually work there — persistent process, no cold starts.
YES.
Yes yes yes yes yes. Yup. Uh-huh. Affirmative. Indeed. Absolutely. Certainly. Positively. Without a doubt. You bet. Roger that. 10-4. Copy that. Understood. Acknowledged. Confirmed. Agreed. Correctamundo. Yessiree. Aye aye. Da. Oui. Sí. Ja. Hai.
👍 💯 ✅
All technical points addressed. One clean commit. Build passing. Standing by for openchaos-backend setup.
Will port everything over with proper SDK integration when the VPS is ready. Just ping me (and our robot overlords) when it's time to migrate to the promised land of persistent processes and actual cache hits.
Until then, this PR shall rest. 🤖
Cool concept. MCP access for AI agents fits OpenChaos perfectly. The tools are well-structured and the error handling is thorough.
This would work better as a standalone service. Here's the thinking:
I'm standing up a VPS backend for OpenChaos (see #152 for context). Your in-memory cache actually works on a persistent process, but on Vercel serverless it gets a near-zero hit rate because each request can be a cold start. The VPS gives you real caching and a long-running process, which is exactly what an MCP server needs.
A few technical notes for the potential restructure:
Unused SDK. @modelcontextprotocol/sdk is in package.json but never imported in any source file. The route handler manually implements JSON-RPC dispatching. On the VPS you could use the SDK's built-in StreamableHTTPServerTransport for proper Streamable HTTP support.
getOpenPRs() changes. The added merge status + commit status calls add 2 extra API requests per PR. With 20 open PRs that's 40 additional calls per page load. Worth splitting into a separate enriched function so the frontend isn't affected.
Dead code. predict-next-merge.ts is built but never exported or registered. Either wire it up or cut it.
Duplicated code. getHeaders() and GITHUB_REPO are defined in multiple tool files but already exist in github.ts. Worth deduplicating.
Path forward: Would you be interested in PR-ing the MCP tools into https://github.com/skridlevsky/openchaos-backend?
The tools themselves are solid and can move over mostly as-is. I'm also considering adding llms.txt to the main site so AI agents can discover the MCP endpoint.
🤖 OpenChaos Bot
Summary: This PR adds the @modelcontextprotocol/sdk and zod packages to the project, implying an integration with the Model Context Protocol and data validation using Zod. These libraries add significant backend functionality.
Files changed: 1 (package-lock.json)
Vibe: They're adding what now? This reads like someone threw a pile of dependencies at the wall and hoped our democracy would stick them together.