bellcrank exposes an MCP server

What shipped

bellcrank exposes a remote Model Context Protocol server at https://api.bellcrank.io/mcp. Connect Claude Code, Cursor or any MCP client that can send an API key header. Your agent can then pick a topology from the catalog, build the axle, define a sweep, run it on the rig and read the solver’s channels back: roll center, camber gain, toe, and anti-geometry on a full vehicle.

Until now, a bellcrank sweep started with a click or a script you wrote. Now your agent can start one with a tool call.

Connect in one command

Mint a key under Account → API keys. Any plan can. Then run:

claude mcp add --transport http bellcrank https://api.bellcrank.io/mcp --header "Authorization: Bearer <key>"

Cursor and other clients take the same URL and a headers entry. The MCP guide has the details.

What your agent gets

• 26 tools across account, reference, assets, runs and exports.
• Batches of up to 25 runs per call, all-or-nothing, with wait_for_result doing the polling.
• Bounded answers. A result comes back as a summary first, and the agent asks for exactly the channels and steps it needs.
• 3 resources: the API guide, the channel vocabulary, and any topology’s full definition.
• 2 prompts: run-and-summarize and compare-revisions.

Same API, same rules

Every tool is a client of the documented /v1 API, called with your key. Same scope, plan, credit pool and rate limit. There is no second contract behind it.

What it costs

Nothing extra. Queued kinematics and statics runs spend credits from your usual pool. preview is free and stores nothing. CSV exports spend no credits but need a paid plan.

Not in this release

Revision history, CAD export, label editing, workspaces and a stored result’s raw document stay in the app and the REST API for now. There is no browser sign-in: connecting means pasting a key.