Limits

MCP has no tier gate — it is available on every plan, including a Free workspace with no card. What bounds it is the same enforcement every surface shares.

Writes stage. They never publish.

schedule_post and reschedule_post create or move a SCHEDULED row, exactly like the composer. There is no tool that publishes immediately — an assistant cannot skip review, and anything it schedules is still visible and editable on your calendar before it goes out.

Plan caps apply, because the tools call the same services

schedule_post goes through the same PostService code path the dashboard and the API use. It inherits:

• The monthly per-platform post cap. Hitting it returns the same POST_QUOTA_EXCEEDED detail the API does.
• Whether your workspace can schedule at all. A Free workspace can write drafts through the dashboard, but MCP has no draft-only tool — schedule_post always attempts a real schedule, so it fails the same way the composer would if you tried to schedule without a card.
• Request rate limits. MCP shares the workspace's programmatic request budget with the CLI and any API key — see Rate limits. A client that loops list_queue is bounded the same way a looping script would be.

There is a separate, tighter limit on unauthenticated MCP requests (30 per minute, keyed by IP) that exists only to bound key-guessing before authentication succeeds — it has nothing to do with your workspace's own budget.

Revoking access

Disconnecting an assistant from AI Assistants in the dashboard deletes its refresh token, so it can no longer mint a new access token. Any access token it already holds is short-lived and expires on its own regardless. Removing yourself from the workspace entirely cuts access at once no matter what — every MCP request re-checks that you are still a member before any tool runs.