MCP (Model Context Protocol)
Mission Control provides an MCP (Model Context Protocol) server that enables AI assistants like Claude to interact with your infrastructure.
Use cases:
- Query the config catalog and explore infrastructure relationships
- List and inspect health checks across your environment
- Execute playbooks to remediate issues or perform operations
- Audit who has access to a resource and when it was last reviewed
- Inspect notification history, playbook run artifacts, and background job runs
- Enable AI-assisted troubleshooting and incident response
MCP Server
Mission Control exposes an MCP server endpoint that AI assistants can connect to for infrastructure interactions.
Available Tools
| Tool | Description |
|---|---|
| Catalog | Query config items and their changes, and explore relationships |
| Access | Audit who has access to a resource, their sign-ins, and access reviews |
| Health Checks | Search health checks, view their history, and run them |
| Playbooks | Run each permitted playbook as its own tool, and inspect past runs |
| Views | Run each permitted view as its own tool |
| Connections | List configured connections, without their settings or credentials |
| Artifacts | Read artifacts produced by playbook runs and health checks |
| Notifications | View the notifications sent for a resource |
| Templates | Evaluate CEL expressions and Go templates |
| Jobs | List scheduled jobs and inspect their run history |
See MCP Tools for each tool's parameters.
Connecting Claude Desktop
Add a custom connector that points at your Mission Control MCP URL. Claude discovers the authorization server and registers with CIMD or DCR.
For client-specific setup, see Client Setup.
Authentication
Mission Control authenticates MCP clients with OAuth. Compatible clients use Client ID Metadata Documents (CIMD). Clients that do not support CIMD fall back to Dynamic Client Registration (DCR).
Point the client at the MCP server URL. The authenticating user needs the mcp:use action. A personal access token used for MCP also needs one of its Roles to allow mcp:use. Each tool then checks the caller's permissions on what it reads or runs, and playbooks need mcp:run. See Permissions.
Example Interactions
Once connected, you can ask Claude questions like:
- "What health checks are failing right now?"
- "Show me all Kubernetes deployments in the production namespace"
- "Run the restart-pod playbook for the api-server deployment"
- "What config changes happened in the last hour?"
Architecture
┌─────────────────┐ ┌─────────────────┐ ┌─────────────────┐
│ Claude/AI │────▶│ MCP Server │────▶│ Mission Control│
│ Assistant │◀────│ (HTTP) │◀────│ API │
└─────────────────┘ └─────────────────┘ └─────────────────┘
The AI assistant decides which tool to call from your request and sends a structured tool call to the MCP server, which Mission Control serves at its /mcp endpoint. Mission Control authenticates the request, checks mcp:use, then runs the tool as you, with the same permission checks as the UI.