PAGGEREDAgent Access · Pro

Bring your own AI to your training.

Agent Access connects an AI client you choose to your Paggered record. It is included with Pro. Start with selected read access; optionally let your agent prepare changes that you review and approve in Paggered.

What you can do

Ask your agent to review the training you actually logged, help prepare a saved routine, or schedule it on specific dates. With the appropriate permissions, it can also propose changes to completed workouts, meals, recovery check-ins, body measurements, goals, equipment setups and fitness preferences.

Pags remains available inside Paggered. Your external AI service has its own subscription, usage costs and data terms; those are not included in Pro.

Connect your client

  1. Use a Paggered account with an active Pro subscription and a client that supports remote MCP with OAuth sign-in.
  2. In your client’s app or MCP connection setup, use the server address below. Follow its connection flow to Paggered.
  3. Sign in to Paggered. Check the client identity and choose which areas it can read: training, nutrition, recovery or body measurements. Connect only a service you trust.
  4. Return to your client and ask it to read a small part of your record, such as your recent workouts. An empty record is a valid result; it is not permission to invent data.
Remote MCP server
https://paggered.app/mcp

ChatGPT and Codex have been verified through the Paggered hosted plugin. Access to custom apps and MCP connections depends on your client, plan and workspace settings. This is not a promise that Paggered is listed in every app directory or that every MCP client works. A separate self-managed connection may require its own setup.

Never paste your Paggered password or a bearer token into a conversation. Sign-in and consent happen on Paggered. Review or disconnect clients in Settings → Connections.

Let an agent propose changes

A connection starts read-only. Ask your agent to check its write context and give you the Paggered permission link. You choose the extra permissions in that form; the agent cannot grant itself access. Existing connections do not gain write permissions automatically.

Creating a record needs the corresponding write permission. Editing or deleting also needs matching read access. Deletion needs a separate permission. If a read area is missing, reconnect with that read permission first.

Permission to propose is not permission to save.

Every batch still needs your separate approval. The agent cannot change your account security, billing or privacy settings, run a live workout timer, or log remote sets one by one during a session.

Review, then approve or reject

For example, ask: “Prepare a saved routine using exercises from my recent workouts. Show me the proposal before saving anything.” This is an example request, not a training prescription.

  1. Your agent prepares a proposal and gives you a Paggered review link. Preparation changes no fitness records.
  2. Open the link and check the account, client, exact changes and any related effects. A proposal can contain up to 50 logical changes, including linked records.
  3. Approve the whole batch or reject it. If anything needs changing, ask for a new proposal. A proposal expires after 30 minutes; a stale or conflicted batch saves nothing.
  4. Use the receipt to confirm what was saved. If a response is interrupted, check the original proposal’s status before asking for another attempt.

Scheduled changes use explicit dates up to 90 days ahead in the reviewed timezone. A request for a recurring plan must be expressed as individual dates within that window.

History, undo and disconnecting

Open Agent activity to see proposals and saved results. An eligible committed batch can be undone as a whole for up to 30 days. You must explicitly confirm undo in Paggered; your agent cannot do it for you.

Undo checks every affected record and dependency. If something has changed since the batch was saved, a conflict can prevent the entire undo. Nothing is partially restored; request a new corrective proposal instead.

You can turn off write permissions or disconnect an agent in Settings. History and eligible undo remain available after disconnection or Pro cancellation. Disconnecting stops future access but cannot recall copies already received by the external service.

Full recovery data is kept for 30 days. Minimal receipts remain with your account after that window; the original health values are no longer available for undo.

If something is missing

  • Only read tools appear: refresh the client’s Paggered tool definition or reopen it. Then ask it to check write context. Refreshing does not grant permissions.
  • Writes are unavailable: check your Pro subscription, connection and selected permissions. The write service may also be temporarily disabled.
  • A save is uncertain: open the original status link. Do not repeat the request with a new command just because a response was lost.
  • A proposal expired or conflicted: nothing was saved by that attempt. Read the latest record and prepare a new proposal for review.

If you still need help, use Paggered support with the request ID and what failed. Do not include tokens or private health records.

Building an integration

MCP and REST share the same operations and permission checks. Use OAuth for Agent Access; the separate TRENBot bearer and Discord channel headers do not authorise these routes.

Write tool → REST operation
get_write_context → GET /v1/agent-access/writes/context
get_write_record → GET /v1/agent-access/writes/records
prepare_write_proposal → POST /v1/agent-access/writes/proposals
get_write_proposal → GET /v1/agent-access/writes/proposals?command_id=…

Read current revisions before editing. Prepare with version agent-access-write.v1, a stable UUID command_id and actions from the generated schema. Give the athlete the returned approval URL, then look up the original command for its receipt. There is no bearer approval or undo endpoint.

After a lost response, query the original command first. An identical replay returns its original result; the same command with a different body is refused. Do not silently retry or issue a new command to bypass uncertainty. Respect weighted quotas and Retry-After; responses are private and must not be publicly cached.

Open the Agent Access API contract