MCP.
What it is.
MCP (Model Context Protocol) is the standard socket between an AI assistant and a data source. Kimbia is one: you bring your own assistant, it queries your activities where they live. This is not a chatbot inside Kimbia - it is your agent, on your data.
Connect.
One address to hand your client, over HTTP transport:
Claude Code
One line in your terminal, available in every project afterwards.
claude mcp add --transport http kimbia https://kimbia.app/mcpClaude (desktop & web)
Settings → Connectors → Add custom connector, then paste the address.
https://kimbia.app/mcpOther clients
Any client that speaks MCP over streamable HTTP works. Pick that transport and paste the address.
https://kimbia.app/mcpOn the first call your browser opens a Kimbia page: sign in with your atproto handle, approve, done. No password, no token to copy around.
What the agent can read.
- Activities, laps, gear
- Every session as your watch recorded it, down to the lap and the split - with filters, sorting and aggregates, so the agent can answer without pulling everything.
- Physiological profile
- HR zones, VDOT / VMA, FTP, weight, target paces, reference performances. The agent prescribes off your real numbers instead of guessing them from past runs.
- Plans and objectives
- Planned sessions and target races, so a week reads inside its cycle rather than on its own.
Tell your agent to start with the kimbia_mcp_describe tool: it returns the full catalogue, the units, and the athletes it is allowed to query.
Ready-made prompts.
The server ships its own prompts too: in a client that supports them (Claude Code exposes them as commands), you pick one instead of describing the task. Arguments complete from your real data - the name of your next race, the athlete you coach.
- prepare_objective
- Where you stand against a race, what the remaining weeks should hold, and what to watch.
- training_review
- A period reviewed: volume and load per sport, races, records, and what the period actually says.
- athlete_brief
- For coaches: a brief on one athlete before a session or a call.
Read-only, for now.
The agent reads, it does not write: nothing in your journal can be changed, created or deleted. Writes are coming, with their own consent step. The one exception, off by default: an anonymous session feedback tool you can enable in Settings → AI / MCP.
Coach and athlete.
A coach only sees an athlete's data while the athlete has opened an access window from their Crew page - from 1 hour to 1 year, revocable in one click. With no open window, the athlete is simply invisible to the coach's MCP.
Revoke.
Every connected client is listed in Settings → AI / MCP, with its last use. One click on “Revoke” cuts its access immediately.
My connected clients