Goal
Turn CT216/mcp2cli into the canonical centralized MCP hub for agents, replacing random local MCP service configs where possible.
Motivation
Current state appears split: Bilby-local mcp2cli has open-brain/fleet-board/n8n configured, while CT216's health endpoint is up but reports no active services. That undermines consistency, safety, auth, and operational observability.
Desired state
- CT216 owns the canonical service registry for shared MCP services.
- Agents use remote mcp2cli clients for shared services instead of local ad-hoc MCP processes.
- Local MCP usage is limited to explicit dev/test fallback.
- Service config, auth, runtime user, logs, metrics, and health are documented and observable.
Work items
- Inspect and document CT216 service unit, runtime user, config path, auth source, and logs.
- Fix
/health so it reports configured/available services, not just active connections.
- Add
/services, /tools, /schema, and /call remote API consistency checks if not already present.
- Add remote-client bootstrap for agents.
- Add read/write safety policy per service.
- Add CI/systemd smoke tests for the LXC deployment.
Acceptance criteria
- CT216 health/status proves configured services and live connectivity.
- Agents can call Open Brain, n8n, and Fleet Board through the hub.
- Runtime is non-root.
- Secrets are not printed in logs or responses.
- Docs define when local fallback is allowed.
Goal
Turn CT216/mcp2cli into the canonical centralized MCP hub for agents, replacing random local MCP service configs where possible.
Motivation
Current state appears split: Bilby-local mcp2cli has open-brain/fleet-board/n8n configured, while CT216's health endpoint is up but reports no active services. That undermines consistency, safety, auth, and operational observability.
Desired state
Work items
/healthso it reports configured/available services, not just active connections./services,/tools,/schema, and/callremote API consistency checks if not already present.Acceptance criteria