If you are preparing this service for real agent workloads, run these checks in order.

​
Fast path: run the bundled check script

npm run reliability:check
Custom server URL: 
npm run reliability:check -- http://localhost:3100/mcp

​
1) Process and transport health

curl http://localhost:3100/healthz
Expected response: 
{"status":"ok","service":"mcp-browser-server","transport":"http"}

​
2) Session lifecycle integrity

npm run agent -- --server-url http://localhost:3100/mcp --sequence \
  browser_start browser_get_info browser_end
You should see a valid sessionId, then a clean close message.

​
3) Navigation and content extraction

npm run agent -- --server-url http://localhost:3100/mcp --sequence \
  browser_start "browser_navigate?url=https://example.com" \
  "browser_get_content?selector=body&type=text" browser_end
This confirms the server can read meaningful page text, not just render a screenshot.

​
4) Visual assertion example

Use element-level screenshots to verify selectors and page state. Example Domain heading captured as element screenshot

​
5) Failure behavior worth testing

  • Stop the target CDP browser while LAUNCH=false and confirm your monitoring catches failures.
  • Provide an invalid selector and verify the error is explicit.
  • Re-run the same deterministic flow repeatedly to detect flaky selectors.