Production-minded Java MCP server that exposes engineering context through typed, inspectable and non-destructive tools.
Open the full-resolution screenshot · Recruiter demo: connect an MCP client with bearer authentication, enumerate typed tools, inspect service context and preview a deployment without mutation.
▶ Watch the recorded authenticated demo (5 seconds)
Deployment status: tested, containerized and ready for an AWS deployment; no public production endpoint is claimed yet.
| Tool | Safety | Purpose |
|---|---|---|
listServices |
Read-only | Discover services and ownership |
serviceHealth |
Read-only | Inspect configured service metadata |
findRunbook |
Read-only | Retrieve incident guidance |
previewDeployment |
Dry-run | Plan deployments without mutation |
estimateMonthlyCost |
Pure | Calculate illustrative compute cost |
The server uses the Spring AI MCP WebMVC starter and publishes real MCP tool callbacks. Remote MCP endpoints require Authorization: Bearer <CONTEXTBRIDGE_API_KEY>; health and the inspection landing page remain public. Supply the production service catalogue through CONTEXTBRIDGE_CATALOG_JSON; safe unconfigured bootstrap entries are used when it is absent.
Set a random CONTEXTBRIDGE_API_KEY of at least 24 characters, then run mvn spring-boot:run.
Open http://localhost:8084 for the inspection dashboard. Configure an MCP client against the WebMVC MCP transport exposed by Spring AI.
- no tool mutates infrastructure
- production deployment previews always require human approval
- unknown resources return structured errors
- tool descriptions explicitly state side-effect behavior
- configuration-backed catalogue prevents hard-coded internal infrastructure data
mvn verify
- OAuth 2.1 resource-server mode for multi-tenant deployments
- MCP resources for runbooks and service schemas
- per-tool authorization policies
- OpenTelemetry tool-call traces
- signed audit events and rate limits
Apache-2.0. Never commit credentials or real internal service data.
