MCP overview

Soriku speaks Model Context Protocol, the standard that lets editors and other tools call AI services through a stable interface. One MCP server exposes Soriku's chat, agents, embeddings and health checks to any MCP-aware client.

Why MCP

Before MCP, every editor wrote its own integration with every AI service. With MCP, editors implement the protocol once and any MCP server (Soriku included) becomes usable without per-editor work. Cursor, VS Code with the MCP extension, Claude Desktop, Zed and others all consume MCP servers.

What Soriku's MCP server exposes

• soriku_chat — send a prompt, get a response routed through smart routing
• soriku_agents — list and pick agent personas
• soriku_models — list available models, local and remote
• soriku_embed — get embeddings for text
• soriku_context — pull project context into the conversation
• soriku_health — server status, useful for verifying the connection

Transport

Stdio transport. The client launches python mcp_server.py as a subprocess and communicates over stdin/stdout. No HTTP, no ports, no firewall config. The MCP server reuses the same engine as the HTTP server, so capability map, agents and routing decisions stay consistent.

Next: setup →