mcpField
H
Hevy Coach

Hevy Coach

Remote
github.com/gCuadrosHealth1Updated 12w ago
Repository

About Hevy Coach

Hevy Coach is an MCP server for Hevy users who want workout analytics and routine management. It calculates metrics from live workout data and leaves workout history untouched.

Key features

  1. 01Live workout and routine retrieval
  2. 02e1RM trends and personal records
  3. 03Volume, consistency, and period reports
  4. 04Bodyweight trends and measurements
  5. 05Routine and folder management

Use cases

  1. 01Review weekly training volume by muscle group
  2. 02Compare workout counts across date ranges
  3. 03Track exercise progress and personal records
  4. 04Create routines or log body measurements

MCP support

Exposes Hevy data and analytics through read tools, resources, and training prompts for MCP clients. Write tools can create or update routines and log measurements; workout history is never modified.

Configuration

Remote servers need no local install. Pick your client, then copy the config or install in one click.

Endpointhttps://hevy-mcp-alpha.vercel.app/mcp
Copy install command
Claude CodeTerminal
Cursor.cursor/mcp.json
VS Code.vscode/mcp.json
Claude Desktopclaude_desktop_config.json
Windsurf~/.codeium/windsurf/mcp_config.json

Run commands in a terminal; paste JSON into the client’s MCP config and restart. Servers that need sign-in will prompt on first connect.

Packages
npmhevy-coach-mcp0.4.2

Tools

Details

From the project README

hevy-coach-mcp

An MCP server for Hevy (the workout tracking app). It fetches your workouts, routines and exercise templates live from Hevy — no local cache or database — then does the analytics math (e1RM, PRs, volume, consistency, period comparisons) so your MCP client can reason over real numbers instead of guessing. It writes back what you ask it to — routines, folders, a bodyweight — and never touches your workout history.

Requires Hevy PRO and a Hevy API key (Hevy app → Settings → API).

Install

claude mcp add hevy -e HEVY_API_KEY=your_key_here -- npx -y hevy-coach-mcp

Or add it manually to your client's MCP config:

{
  "mcpServers": {
    "hevy": {
      "command": "npx",
      "args": ["-y", "hevy-coach-mcp"],
      "env": { "HEVY_API_KEY": "your_key_here" }
    }
  }
}

This works in Claude Desktop (claude_desktop_config.json), Claude Code (.mcp.json or the command above), Cursor (.cursor/mcp.json), and Windsurf (~/.codeium/windsurf/mcp_config.json) — same shape everywhere.

Your API key stays local: it's read from the HEVY_API_KEY environment variable and never leaves your machine except in calls to Hevy's own API.

There's also a hosted mode over Streamable HTTP with OAuth, for clients that can't spawn a local process (Claude.ai, ChatGPT). Point them at https://hevy-mcp-alpha.vercel.app/mcp and they'll run the OAuth flow themselves. docs/CONNECTOR.md covers both setups client by client, plus self-hosting, example questions and privacy details.

First use

Run health-check any time to confirm the connection. Every other tool fetches live from Hevy — there's nothing to sync or warm up first.

Tools

Fifteen read tools, all marked readOnlyHint:

  • health-check — connection status
  • get-workouts, get-workout — list/inspect workouts
  • list-routines, get-routine — list/inspect routines
  • list-routine-folders — the folders routines are organised into, with their IDs
  • search-exercises, get-exercise-history — resolve an exercise by name and see its logged history
  • get-progress, get-records — estimated-1RM trend and PRs (1/3/5/8RM) per exercise; get-progress can also report each session against the bodyweight you were carrying
  • get-volume-report — effective sets and tonnage per muscle group per week
  • get-consistency — training frequency, streak, longest gap
  • compare-periods — volume/workout-count diff between two date ranges
  • get-body-measurements — logged bodyweight and measurements, newest first
  • get-bodyweight-trend — weight change over a range: total, percentage and rate per week

Four write tools, declared as writes so your client asks first:

  • create-routine — build a new routine from exercise names, optionally into a folder named rather than numbered
  • update-routine — edit an existing routine; passing an exercise list replaces the old one outright (destructiveHint)
  • create-routine-folder — add a folder; a title that already exists is never created twice
  • log-body-measurement — record a bodyweight or measurement for a date you give it; re-logging a date keeps whatever else was stored that day

Resources

hevy://profile, hevy://routines, hevy://exercises, hevy://stats/summary, hevy://workouts/recent — cheap live snapshots for a client to read without a tool call.

Prompts

weekly-review, program-audit, deload-check, prepare-session — guide the client through calling the right tools for common training questions.

For the layer above those — a standing system prompt that turns the numbers into decisions — see docs/COACH-PROMPT.md.

Privacy

Nothing about your account is stored — every tool call fetches fresh from Hevy's API using your own key, and nothing about your account is sent anywhere else.

Four tools write to Hevy: create-routine, update-routine, create-routine-folder and log-body-measurement, all declared as writes so your client asks before running them. Nothing writes to your workout history — that is the line, and it does not move: a logged workout is what every number here is computed from, so it can only be changed by you, in the app. Hevy's API has no delete endpoint, so nothing this server creates can be removed from here either. Note that a Hevy API key has no scopes — it grants full account access no matter what this server chooses to do with it.

Contributing

Clone, yarn install, and copy .env.example to .env.local. Validate any change with yarn typecheck && yarn test && yarn build — the same three commands CI runs.

AGENTS.md is the full guide to the architecture, conventions and the decisions worth knowing before changing anything. It is written for AI coding agents, but it is the fastest way for a human to get oriented too.

License

MIT

Comments

Markdown supported

New to MCP?

From what the protocol is to installing your first server, these answers cover most first questions.

An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.