MCP Server

The MCP server is the outward face of matic: it exposes platform capabilities and registered CLI commands as MCP tools so any MCP-compatible client — IDE, agent runtime, or external orchestrator — can drive a matic org programmatically. The pages below split that surface into the layers that matter at integration time: how the server starts and binds a transport, how it negotiates and advertises capabilities, how sessions and per-request context are scoped to an org, how platform capabilities become tool definitions, and how invocations are dispatched onto the platform command envelope. Reach for this section when you need to embed matic into another agent system or reason about what an external caller can and cannot reach.

Startup and Transport

How the MCP server process is brought up, attaches to the daemon, and binds its transport channel before accepting any client.

Capability Negotiation

The handshake that advertises matic's MCP server capabilities to a connecting client and pins the protocol features both sides will use for the session.

Session Lifecycle

How client sessions are opened, tracked, and torn down, and what state the server keeps between requests on a single connection.

Context Scoping

How each session and request is bound to an org, identity, and execution scope so exposed tools resolve against the correct matic surface.

Tool Registration

How platform capabilities and CLI namespaces are projected into the canonical MCP tool inventory served to clients.

Request Dispatch

How inbound MCP tool calls are validated, mapped onto the platform command envelope, executed, and returned as normalized MCP responses or errors.