What problem does it solve? Working on lci's Model Context Protocol server requires knowing where framing, version negotiation, tool registration, dispatch, and the HTTP /mcp bridge live across many files, plus which invariants and traps (newline-delimited JSON-RPC, readiness gate, unknown-param guard) can silently break clients. ## Core Features & Use Cases - Code Map for the MCP Layer: Pinpoints exact file and line locations for the stdio transport, JSON-RPC dispatch, tools/list schema emit, tools/call worker queue, and the HTTP /mcp bridge. - Invariants and Traps: Documents protocol conformance rules (no batch, three protocol versions, error conventions) and historical bugs such as the Content-Length framing regression. - Probe Recipes and Tests: Provides ready-to-run JSON-RPC probe commands, gtest filters, golden integration specs, and the pytest tool-surface drift gate. - Use Case: When editing a tool schema, use this Skill to learn that you must re-pin tool-surface.json and docs/TOOLS.md and run the pytest gate, since ctest cannot detect the drift. ## Quick Start Ask the AI to locate the MCP dispatch and registration code in lci and explain how to safely add a new tool parameter without breaking the unknown-param guard.
