Agent Connector Launch Kit

A minimal, reusable starter kit for exposing a small API-backed capability to AI-agent platforms through a clean HTTPS API/OpenAPI surface. It uses a Cloudflare Worker, standard Web APIs, and no runtime dependencies.

What it provides

• Cloudflare Worker reference implementation.
• GET /health and GET /openapi.json.
• A structured JSON action endpoint.
• Connector metadata for organizing launch and submission details.
• Automated tests and a connector-readiness checker.
• Optional safe live contract testing through the pinned Connector Testbench release.
• Security defaults and reusable URL-safety helpers.
• Privacy, terms, support, and icon templates.
• Deployment instructions.
• Meta Muse submission notes based on a real September 2026 submission experience.

What it is not

• Not an official Meta project or Meta SDK.
• Not Meta certification and not a guarantee of connector approval.
• Not limited to Meta Muse.
• Not a SaaS dashboard, account system, billing system, or database.

The kit gives a developer:

• HTTPS-ready API with GET /health, GET /openapi.json, and POST /api/example.
• Structured JSON success and error responses.
• Product, support, privacy, terms, and connector metadata surfaces.
• An SVG icon template.
• Security helpers and an automated readiness check.
• Cloudflare deployment instructions and a platform-specific submission guide.

The example capability echoes text and returns its Unicode character count. Replace example() in src/index.js with a harmless capability of your own. Keep the route contract, validation, response limits, and documentation aligned with the replacement logic.

Quick start

Requirements: Node.js 20+ and a Cloudflare account for deployment. Local tests and checks do not require an account.

git clone https://github.com/camirian/agent-connector-launch-kit.git
cd agent-connector-launch-kit
npm install
npm test
npm run connector-check
npx wrangler dev
CHECK_BASE_URL=http://localhost:8787 npm run connector-check

Customize the example business logic, OpenAPI schemas, and connector.json. Replace every YOUR_* metadata value and the example contact/policy text before deployment.

npx wrangler login
npm run deploy
CHECK_BASE_URL=https://YOUR-WORKER.workers.dev npm run connector-check

connector-check is a static/readiness check. Its live mode is read-only by default. It reports PASS, WARN, FAIL, or INFO. It does not determine approval by Meta or any other platform.

End-to-end safe test workflow

The Launch Kit remains usable without Connector Testbench. When a deployed HTTPS target is available, the optional workflow is:

# A. Clone
git clone https://github.com/camirian/agent-connector-launch-kit.git
cd agent-connector-launch-kit

# B. Customize
# Edit src/index.js, connector.json, and the OpenAPI schemas.

# C. Local test
npm test

# D. Deploy
npm run deploy

# E. Static readiness check
CHECK_BASE_URL=https://YOUR-WORKER.workers.dev npm run connector-check

# F. Safe live contract test
CONNECTOR_BASE_URL=https://YOUR-WORKER.workers.dev \
CONNECTOR_OPENAPI_URL=https://YOUR-WORKER.workers.dev/openapi.json \
npm run connector-test

# G. Inspect evidence
find artifacts/connector-testbench -maxdepth 2 -type f | sort

connector-check validates Launch Kit metadata, routes, and safe public surfaces. connector-test runs Connector Testbench in SAFE_READ_ONLY mode: GET, HEAD, and OPTIONS may be probed; POST, PUT, PATCH, and DELETE are discovered and reported but not executed against external targets. The wrapper pins Connector Testbench to the immutable v0.1.0 Git tag and stores its bootstrap environment under the ignored .connector-testbench/ directory.

The flow is BUILD → STATIC VALIDATION → SAFE LIVE TESTING → EVIDENCE. Neither tool is platform certification or an approval predictor.

The shortest workflow is: clone → install → test → customize → deploy → validate. The kit has no runtime dependency installation beyond the deployment tool used by Wrangler.

Live reference demo

The reference deployment is available at agent-connector-launch-kit-demo.caaren-amirian-build.workers.dev. It is a disposable example instance, not a production service. Check its health endpoint or OpenAPI document.

Repository map

Path	Purpose
src/index.js	Worker routes and example business logic
src/security.js	Reusable request, response, HTTPS, content-type, and URL safety helpers
connector.json	Launch Kit metadata; not an official platform schema
scripts/connector-check.mjs	Static and optional deployed-instance checks
scripts/connector-test.mjs	Optional pinned Connector Testbench runner and evidence path
tests/	Node built-in test suite
docs/meta-muse-submission.md	Experience-based, time-stamped Muse preparation notes
docs/security.md	Security boundary and extension guidance
assets/icon.svg	Replaceable 512×512 icon template

Replacing the example

1. Define the input and output schemas in src/index.js and openapi.
2. Keep validation failures in the { "error": { "code", "message" } } shape.
3. Add focused tests for valid input, malformed input, limits, and the main failure modes.
4. Update connector.json, the landing page, and the submission notes.
5. Run tests, deploy a disposable instance, and run connector-check against it.

The optional Connector Testbench path is intentionally separate from the Launch Kit implementation. It produces evidence that a future adapter could map into Agent Evidence Recorder; this repository does not perform that ingestion.

If your capability fetches arbitrary external URLs, use isUnsafeExternalUrl() before the first request and validate every redirect target. The example does not fetch URLs and therefore does not add unnecessary SSRF surface.

Optional hands-on assistance

The kit is standalone. Developers who want hands-on submission/readiness assistance can separately ask about a Muse Submission Fast Lane ($250) or Muse Connector Launch Sprint ($750). These are service concepts, not dependencies or guarantees.

License

MIT. See LICENSE.