Tally Workout — MCP Server

Remote Model Context Protocol server for Tally, the strength-training app Claude can write to. Plan your training in conversation — it lands in the app, ready to log.

This repo is the public documentation anchor for the server. The server itself is first-party, closed-source, and runs as part of the Tally backend.

Endpoint

https://mcp.tallyworkout.app/api/mcp

Transport: Streamable HTTP
Auth: OAuth 2.1 — PKCE + dynamic client registration. One-click from claude.ai custom connectors (free tier includes one custom connector slot).
Pricing: Free. Bring your own Claude account.

Related MCP server: FitnessMCP

Setup (~2 minutes)

Get the free iOS app: Tally on the App Store
Follow the guide at tallyworkout.app/connect — paste the endpoint URL into claude.ai → Settings → Connectors → Add custom connector, then sign in with your Tally account.

Claude Code:

claude mcp add --transport http tally https://mcp.tallyworkout.app/api/mcp

Tools

Tool	What it does
`get_training_context`	Snapshot of recent training: volume, frequency, exercises, adherence
`get_workout_history`	Full workout history — sets, reps, weights, planned vs. actually performed
`get_planned_workout`	Read a planned day (for read-modify-write edits)
`plan_workout_day`	Write a planned workout for a single day into the app
`plan_workout_range`	Write a whole week (or any range) of planned workouts
`search_exercises`	Search Tally's exercise library

Design notes

Human edits win. Writes use compare-and-swap semantics; anything you changed or logged on the gym floor is treated as truth and never overwritten by the model.
Planned vs. actual is first-class. History payloads show what was planned and what you actually did, so programming advice is grounded in adherence, not aspirations.
Per-user isolation. OAuth-scoped tokens with row-level security; revocable anytime from Claude's connector settings.