Skip to content

Model Context Protocol (MCP)

Model Context Protocol (MCP)

MCP (Model Context Protocol) is an open standard designed to seamlessly bridge AI models with external tools, local data sources, file systems, and databases. AMC WebUI delivers a complete MCP client implementation.


Supported Server Transports

1. Local Process (stdio)

  • Prerequisite: Requires Docker Compose or self-hosted Node backend.
  • Sample Configuration:
    {
    "mcpServers": {
    "filesystem": {
    "command": "npx",
    "args": ["-y", "@modelcontextprotocol/server-filesystem", "/workspace"]
    }
    }
    }
  • Security Flag: Backend must have ENABLE_MCP_STDIO=true explicitly set.

2. Remote Network Service (sse / http-stream)

  • Protocol: Connects via Server-Sent Events (SSE) or long-polling HTTP streams to remote MCP hosts.
  • Backend Requirement: All MCP transports (both local stdio and remote sse / http-stream) are securely bridged through the Node API service (api container or standalone server/) via /api/mcp/*. Pure static frontend hosting (e.g. Cloudflare Pages) requires a companion Node API backend and cannot run MCP independently.
  • Private Network Access: To allow the API server to connect to intranet or localhost HTTP MCP servers, explicitly enable ENABLE_MCP_PRIVATE_HTTP=true on the server.

Key Management Tabs

Navigate to Settings -> MCP & Extensions:

  1. MCP Tools:
    • Automatically synchronizes tool schemas from active servers.
    • Human-in-the-loop Approvals: Require user confirmation before executing potentially destructive actions (e.g., file edits or database mutations).
  2. MCP Prompts:
    • Discover and apply server-provided prompt templates to instantly scaffold multi-step tasks.
  3. MCP Resources:
    • Mount server-side data files (file:///... or db://...) into conversational context.
  4. Live Logs:
    • Monitor real-time JSON-RPC request and response payloads for rapid debugging.