Command
Use list_projects({}) to confirm my supplied project host is visible. Use my chosen days value of 7, 30, or 90 for every time-windowed call. Read get_overview({host, days}), get_gap_analysis({host, days}), list_actions({host}), and get_agent_analytics({host, section: "crawlability", days}). Select one supported opportunity, propose a content or technical experiment, and draft acceptance criteria using my product facts. Compare follow-up observations only against my saved baseline export. Maintain a proposed decision log in the response, distinguishing saved actions from new recommendations. Execution, prompt editing, publishing, and scheduling require my independent systems and approval; do not claim an autonomous Kept agent. Return evidence links, coverage limits, and explicit unknowns; never infer unsupported scores.What you need
- A visible host, supported window, verified facts, and a saved measurement baseline.
- A human approver and independent execution systems for any implementation or recurring task.
- The external reference skill targets another provider; use the Kept prompt above or adapt its tool calls before running it.
Define a bounded experiment
Combine visibility, citation gaps, saved actions, and crawlability observations to choose one feasible intervention. State the hypothesis, owner, approval gate, and measurement criteria explicitly so the growth cycle remains grounded in observable evidence rather than promises of automatic optimization.
Close the review loop with evidence
After your team executes the approved experiment through its own tools, compare a later supported window with the supplied baseline. Record observations, uncertainties, and the next decision in a review log without claiming that the MCP performed changes or established causal impact.