Sentinel Scan

InspiratEvery engineering team has the same blind spot: nobody knows if a service is

actually secure, fast, and contract-compliant until something breaks in production. As a QA/SDET engineer, I've spent years manually checking this endpoint by endpoint. Sentinel Scan turns that manual check into a one-line Slack command anyone on the team can run, anytime.ion

What it does

Sentinel Scan is triggered by /sentinel . On demand, it runs a live health check covering three categories:

• API contract: hits key endpoints and verifies expected status codes
• Security: checks for missing security headers, confirms protected endpoints reject unauthenticated requests, confirms error responses don't leak stack traces
• Performance: measures p50/p95 latency against a threshold

It also searches the workspace for past messages mentioning the target service in bug/incident/outage contexts, so the report carries institutional memory. Results post back as a Block Kit card with a clear Pass/Warning/Fail verdict, right in the channel where the team already works.

How we built it

• Slack Bolt (Python) handles the /sentinel command and posts Block Kit results
• Playwright's Python APIRequestContext runs the actual health checks
• A custom MCP server exposes the health-check suite as a run_health_suite tool, connected over stdio - making the check engine reusable outside Slack
• Slack's Real-Time Search pulls live in-workspace incident context
• Verdict logic summarizes raw pass/fail data into a short explanation

Challenges we ran into

Getting the MCP stdio transport fully stable under time pressure was the main challenge - subprocess output pollution on the JSON-RPC stream caused intermittent connection issues. The health-check engine was also made directly callable so the agent has a guaranteed-working path while the MCP interface continues to be hardened.

Accomplishments that we're proud of

A fully working, end-to-end Slack agent that runs real, live checks against a real target and reports genuine findings - combining MCP server integration and Real-Time Search to solve an actual daily pain point for engineering and QA teams.

What we learned

New tool usage

What's next for Sentinel Scan

• Finish hardening the MCP stdio connection
• Auto-discover endpoints from an OpenAPI spec
• Add scheduled/recurring scans with regression alerting
• Expand the security check suite with more OWASP-aligned probes

Built With

• anthropic
• mcp
• playwright
• python
• slack-api
• slack-bolt