mcp-servers

MCP Server Management

Manage MCP (Model Context Protocol) servers that provide external tools to the agent. MCP servers extend the agent's capabilities by exposing tools over stdio, SSE, or streamable-HTTP transports.

Agent Tools

The agent has built-in tools for MCP management — no sandbox, network, or CLI needed:

• mcp_list — list all configured servers with status
• mcp_add — add a new server (params: name, command, args, transport, url, env, display_name)
• mcp_remove — remove a server (params: name)
• mcp_status — detailed status for a server (params: name)
• mcp_restart — restart a running server (params: name)

These are the preferred way to manage MCP servers from skills. Use them directly.

RPC Methods

For advanced use or direct API access, the full RPC namespace is documented below.

Listing Servers

// RPC: mcp.list

Returns all configured servers with their connection status, transport type, and enabled state.

Adding a Server

Stdio transport (local process)

// RPC: mcp.add
{
  "name": "filesystem",
  "command": "npx",
  "args": ["-y", "@modelcontextprotocol/server-filesystem", "/home/user/documents"]
}

SSE transport (remote)

// RPC: mcp.add
{
  "name": "remote-tools",
  "transport": "sse",
  "url": "https://mcp.example.com/sse"
}

Streamable HTTP transport

// RPC: mcp.add
{
  "name": "api-tools",
  "transport": "streamable-http",
  "url": "https://mcp.example.com/mcp"
}

With environment variables

// RPC: mcp.add
{
  "name": "github",
  "command": "npx",
  "args": ["-y", "@modelcontextprotocol/server-github"],
  "env": { "GITHUB_PERSONAL_ACCESS_TOKEN": "ghp_..." }
}

With custom display name

// RPC: mcp.add
{
  "name": "gbrain",
  "command": "gbrain",
  "args": ["serve", "--mcp"],
  "display_name": "GBrain Knowledge Base"
}

Returns { "ok": true, "name": "<final_name>" }. If a server with the same name already exists, a numeric suffix is appended automatically (e.g. github-2).