Skip to content

danielopol/claude-code-playbooks

v1.0.0

Search a library of ready-made instruction templates for AI agents and retrieve the one that fits your task.

ChatGPT plugin

Packages the MCP server at https://www.claudecodehq.com/api/plugin/mcp for the plugin directory shared by ChatGPT and Codex. This endpoint has the same tools as /api/mcp, but only serves the playbooks in src/content/plugin-catalog.json.

PathPurpose
plugin.jsonListing details and review test cases
mcp.jsonPoints to the MCP server. It carries no auth block, so choose No authentication in the portal
skills/find-playbook/Tells the model how to search, then show a template
assets/Listing icons, cropped from public/logo.png

Build the zip, which is written to dist/:

npm run plugin:build

The tools themselves live in src/lib/mcp.ts. OpenAI rescans the server daily, so server changes need no new zip. Listing or skill changes do.

What the plugin serves

npm run license:audit regenerates the catalog and research/license-audit/playbook-licenses.csv, whose plugin column gives the decision and reason for each playbook. New playbooks stay out of the plugin until the audit has passed them. To be included, a playbook needs:

  • a permissive license, read from a license file in its source repo
  • a source link that still works
  • no conflict between that license file and the source file's own license: line
  • no Apache NOTICE file in the repo, since those carry extra requirements

The build fails if the listing claims more playbooks than the catalog has.

Submitting

Requirements are from Upload and submit your plugin and Plugin guidelines. Check them again before submitting, because they change.

  1. Deploy the site so /api/plugin/mcp, the destructiveHint annotation, /support, the privacy policy update and the challenge route are live.
  2. Try it privately. In ChatGPT's Plugins page, choose Add custom MCP server, then Create as a plugin, and run every test case in plugin.json.
  3. Verify your organization at platform.openai.com/settings/organization/general. Publishing under an unverified name is rejected. Set developerName and author.name in plugin.json to the verified name, then rebuild.
  4. Upload the zip at platform.openai.com/plugins: Upload new or existing plugin, choose the developer identity, then Upload plugin.
  5. Connect the MCP server with URL https://www.claudecodehq.com/api/plugin/mcp and No authentication.
  6. Verify the domain. Copy the token from the portal into OPENAI_APPS_CHALLENGE_TOKEN in Vercel and redeploy. Confirm that https://www.claudecodehq.com/.well-known/openai-apps-challenge returns only the token, then finish the check and run Scan Tools.
  7. Record a demo that shows the test cases running in ChatGPT, and add the link as the demo video in the portal.
  8. Choose countries in the portal if asked. publication.countries is left out of plugin.json on purpose.
  9. Submit for review and complete the policy attestations. Feedback arrives by email, and you choose when to publish after approval.

No reviewer credentials are needed because the server has no sign-in.

Keep out of the plugin

OpenAI's guidelines say plugins "must not serve advertisements". Keep sponsor cards and affiliate links out of tool responses, and keep pricing and promotions out of the listing text.