MCP server for AI agents
- Context
- In-house product, in production since 2026.
- Starting point
- AI agents need access to systems to be useful. A blank SSH or API-token grant is not a sound foundation for that.
- Role
- Concept, implementation, and operations.
- Outcome
- A modular MCP server in TypeScript with OIDC authentication, graded permission levels, and a discovery flow so an agent learns available operations per host instead of issuing arbitrary commands.
