Physics MCP Server

--- title: Architecture kind: explanation header_svg: src: "/assets/svg/distributed-collaboration-hero.svg" static: "/assets/svg/distributed-collaboration-hero-static.svg" title: "Architecture Overview" animate: true theme_variant: "auto" reduced_motion: "auto" --- # Architecture <p align="center"> <img src="assets/header.svg" width="960" alt="Physics MCP banner" /> </p> [Home](../README.md) · [Architecture](Architecture.md) · [Configuration](Configuration.md) · Tools: [CAS](Tools/CAS.md) · [Plot](Tools/Plot.md) · [NLI](Tools/NLI.md) - Server: TypeScript MCP server orchestrates CAS, Plot, and NLI tools. - Transport: JSON-RPC over stdio compatible with MCP clients. - Python worker: Performs CAS and plotting, returns results to server. - NLI: Optionally uses a local LM API (e.g., LM Studio) when configured; otherwise falls back to a rule-based parser. Flow - MCP client sends `tools/list` and `tools/call`. - Server registers tool metadata from tool packages and routes calls by prefix (`cas.*`, `plot.*`, li.*`). - CAS/Plot calls are proxied to the Python worker via a lightweight JSON-RPC protocol over stdin/stdout. - NLI calls go to a local LM REST API (`/chat/completions`) when configured; otherwise a rule-based parser is used. Key Files - Server entry: `packages/server/src/index.ts` - Python worker: `packages/python-worker/worker.py` - CAS tools: `packages/tools-cas/src/index.ts`, `packages/tools-cas/src/schema.ts` - Plot tools: `packages/tools-plot/src/index.ts`, `packages/tools-plot/src/schema.ts` - NLI tools: `packages/tools-nli/src/index.ts`, `packages/tools-nli/src/schema.ts` - Local MCP types (fallback): `packages/mcp-types/` Notes - The server tries to import the official MCP SDK; if missing, it uses a local mock implementation. - The Python worker uses SymPy, NumPy, SciPy, Pint, and Matplotlib. - Images are returned as base64-encoded PNG; CSV data is included where relevant. Humorous invariant: this diagram conserves coherence better than most wavefunctions.