Why MCP matters
Without MCP, every new tool requires a custom integration. With MCP, any tool that speaks the protocol is immediately available to MIRA:- No custom code — add a server URL or command, and the tools appear automatically
- Namespaced — each server’s tools are prefixed to avoid collisions
- Safe — MIRA prompts you before executing any tool the first time
How it works
- MIRA connects to the server and calls
list_toolsto discover available tools. - Tool definitions (name, description, JSON schema for inputs) are injected into the engine’s context.
- When the engine decides to use a tool it calls
call_toolwith validated arguments. - The server executes the tool and returns the result as text or structured data.
Transport types
MIRA supports two MCP transports:| Transport | Description | Best for |
|---|---|---|
| stdio | MCP server runs as a subprocess; MIRA communicates via stdin/stdout | Local CLI tools, scripts |
| SSE | MCP server runs as an HTTP server; MIRA connects via Server-Sent Events | Remote APIs, shared team servers |
MCP tool namespacing
Tools are prefixed with the server name to prevent collisions:Next steps
Add a stdio server
Connect a local CLI-based MCP server
Add an SSE server
Connect a remote HTTP-based MCP server
Edit this page — Open a pull request