RankPilot exposes its actions to AI assistants through the Model Context Protocol. The endpoint is /api/mcp on the app origin. Clients that support OAuth register and authorise themselves; other clients use an MCP key from Account, API keys.
Tools
list_actions(read): Open work ranked by expected goal units times calibrated confidence divided by effort.open_investigation(write): Start an investigation of a finding or a question. Asking again returns the one that is open.get_investigation(read): Read an investigation: hypotheses with confidence wording, evidence basis and alternative explanations.create_task(write): Create a task from a finding, an investigation or by hand. Risky kinds wait for a person to approve them.update_task_status(write): Move a task to in_progress or deployed, or comment. Approval, verification and closing stay with people and the engine.get_track_record(read): How often recommendations worked here, with the sample size and the inconclusive count.get_brief(read): A content brief as JSON or Markdown, with the limits of its evidence.
Read tools never change anything. Write tools act as the person who authorised the client, with the same permissions and the same audit log entry as the same action in the app.
Quotas
MCP keys are limited per plan: the Free plan has none, and the plan table on the pricing page lists how many each paid plan includes. Hosted language-model calls follow the plan caps: Starter 50, Pro 300, Agency 1500 a month.
Related
The REST API is documented in the API reference, and event delivery in webhooks.