MCP & agents
Open-FDD supports MCP (Model Context Protocol) for AI-assisted commissioning, validation, and documentation — grounded in the real REST API, not simulated data.
| Guide | Content |
|---|---|
| MCP setup | Images, entrypoints, tools |
| MCP role packs | Four agent contexts: package/mapping, surrogate-train, Unity WebGL, operator (portable ZIP lanes; no Jupyter-in-SPA) |
| Companion — WattLab + EnergyPlus | Twin/ECM: three surfaces, golden loop, dual-MCP |
| Dual-site MCP IT | Liberty B50≠B100 accuracy / historian / findings |
| CSV flood + AFDD routine | BUILDING_50 hourly append + rule patches (agent doc) |
| E+ dump + clustering | Dump zip + pandas sklearn export |
| FDD ops → Twin knobs | Pointer: open-fdd findings → vibe20 G14 dial (no IDF in mcp) |
| Agent safety | Hard boundaries for automation |
| Cursor & OpenClaw | IDE and edge agent wiring |
| External agent workflow | Codex, Cursor, MCP patterns |
Repository reference: mcp/README.md