Your organization administrator must enable MCP Creator Access. The flag is off by default, and you must be an owner, admin, or creator. Reconnect after access is enabled to discover the tools. Each call checks your current access, including when you select another organization. Recommendation planning also requires Recommended Reports.

1. Call get_report_creation_options to see available report types, credit costs, remaining credits, enabled models, and feature flags.
2. Use preview_report_prompts with your brand, domain, optional guidance, and a client-generated previewId to research and inspect prompts. It charges the normal preview credits and saves a reusable plan. Read it later with list_report_plans, get_report_plan, or get_prompt_preview.
3. Call create_report with the chosen report type, brand, domain, and config. Preserve the preview’s complete compiled steering when reusing it. Creation charges the normal report credits and returns report/job IDs.
4. Call get_generation_progress with kind: "report" and the report ID. Follow nextCheckAt instead of repeatedly polling. Read the report when terminal status is completed.

Progress is available for report, simulation, preview, and recommendation. It returns the current state, stage information, an estimated finish time when available, its source, and when to check back next. Simulation progress uses the edit ID and also needs the report ID. Preview progress accepts the preview ID or queue item ID; recommendation progress uses the job ID. Estimates can change or be overdue. An unknown estimate remains unknown. A terminal state stops polling; waiting for human input or approval also stops automatic polling.

A queued preview needs an explicit continuation: wait until the suggested time, then call preview_report_prompts with the same inputs, preview ID, and returned queueItemId. Checking progress alone does not start it.

For source edits, first read get_action_detail and get_page_content. Apply focused, supported improvements to the full Markdown page, then call run_simulation with a stable edit ID. It charges the normal simulation credits. Use progress for check-back guidance and get_simulation_result for measured Perception changes. Cancel an active simulation with cancel_simulation, or an active report with cancel_report_generation; cancellation does not promise a credit refund.

generate_report_recommendations researches and saves recommended report plans. Supply a stable UUID requestId and reuse it on retries. Read progress, get_report_recommendation_job, and list_report_recommendations to inspect the results. Generating plans does not start full reports; explicitly call create_report for the plans you choose.

Report creation is not idempotent: after an uncertain connection failure, inspect list_reports before retrying. An active simulation start rejects duplicate requests, but starting the same edit after completion spends credits again.