Help center

Connect Tempoto your assistant.

Tempo runs an MCP server: a compatible AI assistant can read your plan, your day and your progress there and — if you allow it — draft or adjust workouts in Tempo. It is optional, and Tempo works fully without it.

Back to home

The address to paste

Every client below asks for the same thing: the server address. OAuth discovery and client registration then happen on their own.

Tempo MCP server

https://tempo.zenocode.dev/mcp

OAuth discovery

https://tempo.zenocode.dev/.well-known/oauth-protected-resource/mcp

A client that cannot find the authentication on its own can start from this address. Tempo accepts dynamic registration, so there are no client credentials for you to create by hand.

What you are granting

Connecting goes through a Tempo consent screen: you sign in, then you choose which permissions this client gets. Nothing is shared until you approve.

A coach account additionally sees the coaching tools. On an account without a coach space those tools simply do not exist.

The permissions

training:read: Read your training profile, your plan, your day, your progress and the exercise library.
training:write: Create and adjust your plans, draft workouts, move a workout to another day.
coaching:read: Read your coach space: content, consenting client relationships and assignments. Coach accounts only.
coaching:write: Create or revise your coach content and assign it to consenting clients. Coach accounts only.

Revoking access

Profile → Connections lists every connected client and its permissions. Revoking cuts access immediately: tokens issued before the revocation are refused.

Connect an assistant

ChatGPT adds an MCP server by its address from developer mode, on the paid plans.

Where that setting lives has moved more than once: look for “developer mode” in the settings, near apps and connectors.

In the ChatGPT settings, turn developer mode on.
Add a custom app or connector and paste the Tempo server address.
Choose OAuth authentication with no client credentials: ChatGPT registers itself with Tempo.
Sign in to Tempo and grant the permissions on the consent screen.
Enable the connector in the composer. ChatGPT asks for confirmation before every action that writes.

ChatGPT only reaches a public HTTPS address: a Tempo instance running locally will not work without a tunnel.

Claude connects to a remote MCP server through its custom connectors, on the Free, Pro, Max, Team and Enterprise plans.

Copy the Tempo server address above.
In Claude, on the web or in the desktop app, open settings, then Connectors.
Choose “+”, then “Add custom connector”. On a Team or Enterprise plan the owner adds it from the organization settings instead.
Paste the address and confirm. Leave the advanced OAuth fields empty: Tempo registers Claude on its own.
Hit “Connect”: Tempo asks you to sign in, then to grant the permissions. Approve.
In a conversation, turn the Tempo connector on from the “+” menu.

From the terminal, with Claude Code

One command is enough; then run /mcp in Claude Code to open the Tempo consent screen.

claude mcp add --transport http tempo https://tempo.zenocode.dev/mcp

At Google, MCP support does not move at the same pace everywhere. The path we can describe faithfully is the Gemini CLI; in the Gemini app and in Gemini Enterprise, custom MCP servers are rolling out gradually and depend on the plan — check Google’s own documentation.

Register the server with the Gemini CLI:
gemini mcp add --transport http tempo https://tempo.zenocode.dev/mcp
Start gemini, then run /mcp auth tempo: the CLI sees that authentication is required and opens the Tempo consent screen in your browser.
Run /mcp list to confirm the Tempo tools are there.

Google does have a connector marketplace, but not in the “add any third-party MCP server with one click” sense Claude and ChatGPT offer: Gemini CLI has its own extensions catalog (gemini extensions list, installable from GitHub), and Gemini Enterprise (formerly Agentspace) has a connector gallery for enterprise integrations. Either way, Tempo is still added manually via /mcp add.

When it does not work

The connection succeeds, then everything returns 401: Three possible causes: your email address is not verified, consent was never completed for this client, or access has since been revoked. Verify your email, then reconnect the client from the assistant.
The assistant mentions an insufficient scope: The tool it called needs a permission you did not grant — typically training:write for anything that changes your plan. Reconnect the client and grant it.
The coaching tools do not show up: They exist only for accounts with a coach space. On a trainee account the assistant gets “unknown tool” rather than a permission error: that is expected.
I revoked access and the assistant keeps trying: Tokens issued before the revocation are refused, but the client still has one cached and retries. Remove and re-add the connection in the assistant to start clean.
The client cannot find the server: The address must end in /mcp and be reachable over HTTPS from the internet. A hosted assistant cannot reach a local instance.