Prog Strength MCP
Prog-Strength/prog-strength-mcp
Acesso a dados reais de musculação para agentes IA analisarem.
Ver no GitHubServidor MCP que expõe dados reais de treino de força para agentes IA, permitindo análise baseada em dados concretos em vez de suposições. Integra-se ao Prog Strength, um rastreador de treino de peso nativo de IA, via protocolo HTTP streamable usando FastMCP.
Instalação
npx convez-ai Prog-Strength/prog-strength-mcp1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Análise inteligente de histórico de treinos e progressão de força
- Recomendações personalizadas de exercícios baseadas em dados históricos
- Rastreamento e otimização de rotinas de musculação com IA
Veredito

Solução técnica sólida que resolve um problema real: conectar agentes IA a dados concretos de treino. Arquitetura clara com proxy transparente e modelo de autenticação bem pensado via JWT. Ideal para quem quer criar agentes que raciocinem sobre dados reais de musculação.
README
A tradução para o português deste README ainda está na fila.
prog-strength-mcp
A Model Context Protocol (MCP) server that exposes Prog Strength data to LLMs.
Built with FastMCP and served over
streamable HTTP. Every tool call proxies to the Go Chi API in
prog-strength-api —
this server never reads or writes the SQLite database directly.
Architecture
Claude / MCP client ──HTTP──▶ prog-strength-mcp (FastMCP) ──HTTP──▶ prog-strength-api (Go Chi) ──▶ SQLite
Auth model. This server is a transparent forwarder. The MCP client
(the agent) opens each session with the end-user's JWT in the HTTP
Authorization header; FastMCP exposes that header to tool handlers
via get_http_headers, and the handlers pass it verbatim through to
the API. MCP holds no signing key and cannot mint tokens. The
ability to impersonate users was deliberately removed when the
production frontend went live.
Calls that arrive without an Authorization header are rejected by
the tool handler before they hit the API. Public endpoints (currently
just list_exercises) don't require auth.
Tools
| Name | Description |
|---|---|
list_exercises | Public catalog browse. Optional muscle_group and equipment filters. |
list_workouts | The calling user's logged workouts (capped at 50, most recent first). Identity from Authorization header. |
create_workout | Log a new workout for the calling user. Identity from Authorization header. |
Local development
Requires Python 3.12+ and uv.
# install deps (creates .venv and uv.lock)
uv sync
# the server has no secrets to configure; defaults work for local dev.
# copy the env template only if you want to override PROG_STRENGTH_API_BASE_URL
# or the bind host/port.
cp .env.example .env
# run the server (loads .env automatically if you run via `uv run`)
uv run prog-strength-mcp
The server listens on http://localhost:8000/mcp by default. Point an MCP
client at that URL.
To run against a local API instead of prod:
# in one terminal — the Go API on :8080
cd ../prog-strength-api && go run cmd/api/main.go
# in another — the MCP server proxying to it
PROG_STRENGTH_API_BASE_URL=http://localhost:8080 uv run prog-strength-mcp
Configuration
All config is via environment variables — see .env.example for the full list.
| Variable | Default | Purpose |
|---|---|---|
PROG_STRENGTH_API_BASE_URL | http://localhost:8080 | Where to proxy tool calls. |
MCP_HOST | 0.0.0.0 | Bind address for the streamable-HTTP transport. |
MCP_PORT | 8000 | Bind port. |
No secrets are required — the server holds none.
Deployment
Shares the EC2 host with prog-strength-api. Each service has its own
docker-compose file; they join a shared external docker network called
prog-strength so Caddy (in the api stack) can reverse-proxy to
mcp:8000 and this server can reach the api at http://api:8080.
Caddy serves https://mcp.progstrength.fitness and terminates TLS — the
mcp container has no public ports.
Pipeline: semantic-release in .github/workflows/release.yml tags
and SSH-deploys on feat:/fix: commits to main. The deploy step
SSHes into the EC2 host, git pulls /home/ubuntu/prog-strength-mcp,
writes a minimal .env with just APP_VERSION, and runs
docker compose up --build -d. No app secrets to wire in.
Required GitHub secrets on this repo:
| Secret | Value |
|---|---|
EC2_HOST | Same Elastic IP as the api repo. |
EC2_SSH_KEY | Same private key as the api repo. |
Host prerequisites (handled by the infra repo's bootstrap on fresh hosts; do these once on the existing host):
docker network create prog-strength
git clone https://github.com/Prog-Strength/prog-strength-mcp.git \
/home/ubuntu/prog-strength-mcp
Layout
src/prog_strength_mcp/
├── __init__.py
├── __main__.py # `python -m prog_strength_mcp` and console script entry
├── server.py # FastMCP instance + register() calls + run()
├── workouts.py # list_workouts + create_workout tools
├── exercises.py # list_exercises tool (public)
├── api_client.py # httpx async wrapper, header-passthrough only
└── config.py # env var loading (no secrets)