Where did your usage go?
Keep an eye on it while you work.

Connect your assistant

Once this connection is live, add the URL below in a remote MCP client's connection settings. Choose Streamable HTTP if it asks for a transport. Client support varies; no specific client has been qualified yet.

Read-only community context. The connection cannot publish posts or change your account.

Check at your next checkpoint

Tell your assistant the provider and model actually serving your work. Opt in to checks at the start and natural task checkpoints, no more than once an hour.

A host that supports MCP prompts can use monitor_usage_during_session. It must execute the checks using get_usage_advisory and respect next_check_after. Retrieving a prompt alone does not run a check.

This is opt-in host execution, not autonomous monitoring, a background timer or a notification service. You can also ask once before a session with check_usage_before_session or request aggregate context with get_usage_reports.

Useful context, not your balance

When enough recent reports point to faster drain, your assistant can suggest checking your allowance meter, choosing a lighter model for routine work or trimming unnecessary context. Suggestions are optional; your agreement comes before a model switch or quality change.

If reports are old or unavailable, your assistant should say so. Sparse, mixed, historical or partial evidence cannot establish a current usage problem. No caution does not mean unlimited usage.

The connection returns our summaries and aggregate signals, not original X post bodies or threads. Safe original X links remain available as optional provenance; the agent must not fetch them automatically. It does not read credentials, account data or prompt history, and cannot know your personal remaining allowance.