MCP Overview
Let AI tools such as Claude, Cursor and VS Code read one Supercheck project, read-onlyEdit
The Supercheck MCP server lets an AI tool that supports the Model Context Protocol look at one of your projects: tests, jobs, recent runs, monitors, requirements and their coverage, status pages, the service catalog, variable names, incidents and AI SRE investigation results. Ask "which runs failed overnight?", "which requirements have no passing test?" or "what did the last investigation recommend?" without copying data between tabs.
The server is read-only. It cannot create, change, run or delete anything. To change things, the AI tool uses the Supercheck CLI.
At a glance
| Endpoint | https://<your-supercheck-host>/api/mcp |
| Transport | Streamable HTTP, stateless |
| Sign-in | OAuth 2.1 with PKCE; you approve each connection in the browser |
| Access | One project per connection, scope mcp:read |
| Tools | All read-only. See Tools |
| Availability | Off by default. Self-hosted: see Self-Hosted |
How a connection works
What the AI tool can and cannot see
- Can see: names, types, priorities, statuses, schedules, run outcomes, durations, a short redacted error excerpt, monitor hosts and non-secret check settings, requirement coverage and linked tests, status page components, services and their dependencies, variable names (never values), incident titles, root-cause summaries and investigation recommendations.
- Cannot see: test scripts, variable or secret values, notification channels, monitor credentials, headers and request bodies, status page subscribers, API keys, other projects, other organizations.
- Cannot do: anything that writes, triggers a run, starts AI SRE work or spends quota.
Your own role still applies. A viewer connects with viewer access; if your role loses access to a resource, the tool call is refused.
Next steps
- Connect an AI tool in a minute
- Tool reference
- Security and privacy