MCP Integration

Source: backend/app/mcp/, src/renderer/src/components/Settings.tsx

Overview

Model Context Protocol (MCP) support. Connect to MCP servers, discover tools, execute them, and generate training data from tool schemas.

What is MCP

MCP is a standard protocol for connecting LLMs to external tools and data sources. An MCP server exposes tools (functions with typed parameters) that can be called by an LLM or orchestrator. SiliconDev supports the stdio transport — it launches the server as a subprocess and communicates via stdin/stdout.

Server Management

Managed in the Settings page under "MCP Servers".

Add Server

Provide:

Field	Example	Description
Name	filesystem	Display name
Command	npx	Executable to run
Args	-y @modelcontextprotocol/server-filesystem /tmp	Space-separated arguments

The server config is saved to ~/.silicon-studio/mcp_servers.json.

Test Connection

Click "Test" to connect to the server and list its tools. Shows the number of tools found or an error message.

Remove Server

Removes the config. Does not affect the server binary.

Tool Discovery

GET /api/mcp/servers/{id}/tools connects to the server, initializes a session, calls list_tools(), and returns the tool schemas.

Each tool has:

{
  "name": "read_file",
  "description": "Read the contents of a file",
  "inputSchema": {
    "type": "object",
    "properties": {
      "path": { "type": "string", "description": "File path to read" }
    },
    "required": ["path"]
  }
}

Tool Execution

POST /api/mcp/execute with server_id, tool_name, and tool_args. The backend connects to the server, calls the tool, and returns the text result.

Dataset Generation

In the Data Preparation page, the MCP tab lets you generate fine-tuning datasets from tool schemas. See Data Preparation for details.

Current Limitations

• MCP tools are not directly callable from the standard Chat interface. However, they can be invoked by the agent within the Agent Terminal.
• Only stdio transport is supported (not HTTP/SSE).
• Each tool call creates a new server connection (no persistent sessions).

Backend Architecture

backend/app/mcp/
  __init__.py
  registry.py   -- CRUD for ~/.silicon-studio/mcp_servers.json
  client.py     -- MCP SDK wrapper (StdioServerParameters, ClientSession)
  service.py    -- Orchestrator combining registry + client

The client uses the mcp Python package. Connection flow:

params = StdioServerParameters(command=cmd, args=args, env=env)
async with stdio_client(params) as (read, write):
    async with ClientSession(read, write) as session:
        await session.initialize()
        tools = await session.list_tools()