Remote MCP servers and a Claude plugin for Rush SR owners and anyone who analyzes AiM data. Built by Rush Auto Works.
They work in Claude and in ChatGPT: ask a question about your Rush SR, or give it an AiM RaceStudio lap export or an .xrk log and ask who was faster and where.
| Piece | What it does | Where |
|---|---|---|
| Rush SR maintenance server | Diagnoses a symptom, looks up a service interval, finds the steps for a procedure, all from the public owner's manual, with deep links | servers/maintenance, https://maintenance.mcp.rush.sr/mcp |
| Rush SR lap server | AiM RaceStudio CSV analysis: lap times, best and theoretical best lap, compare two laps, where a lap loses time, compare two sessions or two drivers | servers/laps, https://laps.mcp.rush.sr/mcp |
| Rush SR plugin for Claude | Both servers plus a skill that reads AiM .xrk logs and full-size CSVs in Claude's sandbox |
plugin/ |
| US road courses | 93 road-racing venues with street address, coordinates and whether each is still running; the same list the dealer CRM uses for "Nearest tracks" | data/us-road-courses.csv |
Add either URL as a custom connector in Claude (Customize, Connectors) or in ChatGPT developer mode (Settings, Security and login, Developer mode, then ChatGPT Plugins). No sign-in is needed.
For .xrk files in Claude, zip the plugin/ folder and upload it under Customize, Plugins, Add, Upload plugin. Attach an .xrk file and ask "analyze my laps". The skill converts it with libxrk and runs the same analysis engine as the lap server.
AiM .xrk ─▶ convert.py (libxrk) ─▶ RaceStudio-style CSV ─▶ engine.mjs (node) ─▶ lap analysis
The servers analyze a CSV in memory and store nothing. They are retrieval and arithmetic only and make no calls to a language model. The plugin README lists everything the skill runs and sends.
Each server is its own package.
cd servers/laps # or servers/maintenance
npm ci
npm run typecheck
npm test # boots wrangler dev and runs the E2E suite
npm run build:skill # laps only: regenerates plugin/skills/rush-sr-laps/scripts/engine.mjs
The converter tests need numpy and run with python3 -m unittest discover -s tests. CI fails if engine.mjs drifts from the source.
Design notes are in docs/: platform findings, and a plan with a failure matrix for each server and for the skill.
MIT, see plugin/LICENSE and LICENSE.