Audit, Fix, and Recheck Your Launch With an AI Agent

Your app is built. The useful next agent task is to find one launch gap you can verify. This walkthrough uses LaunchBuddy's public MCP tools to audit a page, keep the evidence, and recheck a change. It describes a reproducible workflow; the example finding below is illustrative, not a fabricated scan result.

1. Start from a public URL

Install the existing LaunchBuddy integration, or use the public audit page. The MCP endpoint is https://launchbuddy.com/api/mcp. Public audit and resource tools do not require an account. Use a public URL you own or are authorized to assess; keep private URLs and credentials out of the request.

For a visual introduction, watch LaunchBuddy scan launchbuddy.com. That recording shows the scan and report flow. It does not show a customer conversion result.

2. Ask for separate checks and save the original response

Use LaunchBuddy audit_my_launch with:
{"url":"https://launchbuddy.com","checks":["launch","technical","agents"]}

Save the complete returned audit JSON as the baseline. Show the separate
scores, the supporting evidence, and every unavailable check. Do not turn
an unavailable check into a pass. Do not combine the scores into a new score.
Choose one finding the response actually supports. Stop before editing,
publishing, connecting an account, or starting paid work.

The URL above is LaunchBuddy's own public site. Replace it with your public URL for your audit. The result can vary as the site changes and as remote checks become available. Keep the actual response rather than copying an example score.

3. Turn one finding into a small change

Suppose the returned evidence says a visitor cannot see the product result before signup. Ask for a short recording of the working product and an explanation beside the signup action. If that finding is absent, choose a different supported finding.

For the finding we selected, call search_launch_resources with a specific
query and the matching gap. Read the chosen result with get_launch_resource
using its returned ID. Give me one proposed edit, its supporting evidence,
and a done-when check. Do not invent a resource ID, testimonial, or result.

Review the change, implement it in your own project, and deploy it through your usual review process. A proposed edit is not a shipped fix.

4. Recheck the live change

Use verify_launch_audit with the same public URL and the complete original
audit JSON in previous_audit. Compare each native finding. List what is
verified, what is still failing, and what remains unobservable. A fresh
score alone is not proof that the edit improved signups or sales.

If a check fails to run, retain that uncertainty and retry when the source is available. Do not mark a task verified because the request returned HTTP 200.

5. Save a project only if you need a durable checklist

The public combined audit is useful on its own. If its response includes an assessment continuation and claim token, preserve that private continuation. Otherwise use assess_launch to obtain a saved assessment and claim token. With the owner's approval, connect in the browser, then call claim_assessment using the returned token. Do not invent a token or repeat a scan solely to save it.

Call create_growth_plan with the returned project ID, then read get_launch_todos. preview_recommended_asset shows the proposed asset without starting provider work. Actual asset generation and payment still require the explicit review shown by the product. Never paste a claim token or private workspace link into a public post.

What to measure

Count saved assessments, connected projects, created plans, completed tasks, previewed assets, requested assets, and verified paid orders separately. An MCP initialization is a protocol event and may come from a directory or health check. A checked task can be self-reported; a verification receipt is stronger evidence. Neither is automatically customer adoption.

For a traffic experiment, use one campaign label across the tutorial and its video. Compare qualified outcomes after the scan, and keep internal demonstrations and QA traffic out of the customer cohort. Start with the free scan if you want to see the public report before connecting anything.