MCP quickstart

An assistant reads and writes a log through the MCP tool profile: its core tools (log, add sets, edit, delete, undo, read, records, progress, search) work the same on every compatible server.

Connect an assistant

Sandbox, with sign-in. Claude Code:

claude mcp add --transport http lifterclub-sandbox https://lifterclub-mcp.runmighty.dev/

Then run /mcp in Claude Code and sign in with Google when asked. On claude.ai, add a custom connector with the same address. Try "log squats 5x5 at 100 kg" and "what did I do this week?".

Sandbox, with a key. Make a personal key on the sandbox's account page, https://lifterclub-api.runmighty.dev/account, and send it as a header:

claude mcp add --transport http lifterclub-sandbox https://lifterclub-mcp.runmighty.dev/ --header "Authorization: Bearer lck_…"

The template server, with its fixed token (see Where to run them):

claude mcp add --transport http ows-template http://localhost:8080/ --header "Authorization: Bearer dev-token"

What an assistant keeps to

The profile's README says what a server does and what an assistant does: log only what the person said, mark what you inferred estimated, ask when a result lists needs_confirmation, tell the person the summary line, keep numbers out for someone in calm mode, and send an idempotency_key with every write.

Building your own server

Start from mcp/template: Go, the official MCP SDK, both cores over one store. Check it, or any server, against the profile:

go install github.com/openexercise/open-workout/cmd/ows@latest
OWS_TOKEN=dev-token OWS_READ_TOKEN=read-token ows conformance mcp http://localhost:8080/
OWS_TOKEN=dev-token ows conformance http://localhost:8080