Skip to main content
Agent Chassis has native support for the Model Context Protocol (MCP), allowing you to connect external data sources and tools seamlessly.

​
What is MCP?

MCP (Model Context Protocol) is a standard protocol for connecting AI agents to external data sources, tools, and services. Agent Chassis supports three transport methods:
  1. Stdio - Local subprocess communication
  2. SSE - Server-Sent Events (remote HTTP)
  3. Streamable HTTP - HTTP-based streaming (remote)

​
Configuration

MCP servers are configured in mcp_config.json: 
{
  "mcpServers": {
    "server-name": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-name"]
    }
  }
}

​
Stdio Transport (Local)

For local Node.js-based MCP servers: 
{
  "mcpServers": {
    "memory": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-memory"]
    },
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "."]
    }
  }
}

​
SSE Transport (Remote)

For remote MCP servers using Server-Sent Events: 
{
  "mcpServers": {
    "remote-server": {
      "url": "https://mcp.example.com/sse",
      "transport": "sse",
      "headers": {
        "Authorization": "Bearer your-token"
      }
    }
  }
}

​
Streamable HTTP Transport (Remote)

For remote MCP servers using HTTP streaming: 
{
  "mcpServers": {
    "Context7": {
      "url": "https://mcp.context7.com/mcp",
      "transport": "streamable-http",
      "headers": {}
    }
  }
}

​
Built-in MCP Servers

​
Context7

Documentation search and library information: 
{
  "mcpServers": {
    "Context7": {
      "url": "https://mcp.context7.com/mcp",
      "transport": "streamable-http"
    }
  }
}
Tools available:
  • resolve-library-id - Find library IDs
  • get-library-docs - Fetch documentation

​
Memory Server

Persistent memory/knowledge graph: 
{
  "mcpServers": {
    "memory": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-memory"]
    }
  }
}
Tools available:
  • create_entity - Create entities in knowledge graph
  • search_entities - Search the knowledge graph
  • update_entity - Update entity properties

​
Using MCP Tools

​
1. Tools are Auto-Discovered

When you start the server, MCP tools are automatically discovered and registered: 
Starting up Agent Chassis...
Loading MCP server: Context7
Connected to MCP server (Streamable HTTP): Context7
Loading MCP server: memory
Connected to MCP server (Stdio): memory

​
2. Use Tools in Agent Requests

Include MCP tools in allowed_tools: 
{
  "messages": [
    {"role": "user", "content": "Search for Python documentation"}
  ],
  "model": "kimi-k2-thinking",
  "allowed_tools": ["resolve-library-id", "get-library-docs"]
}

​
3. Tool Filtering

Restrict which tools are available per request: 
{
  "messages": [
    {"role": "user", "content": "What time is it?"}
  ],
  "model": "kimi-k2-thinking",
  "allowed_tools": ["get_server_time"]  // Only allow this tool
}

​
Hybrid Tooling

Agent Chassis seamlessly mixes MCP tools and local Python tools: 
{
  "messages": [
    {"role": "user", "content": "Calculate 5*5 and search for math docs"}
  ],
  "model": "kimi-k2-thinking",
  "allowed_tools": [
    "calculate",              // Local Python tool
    "resolve-library-id"      // MCP tool
  ]
}

​
Tool Translation

All tools (MCP and local) are automatically translated to OpenAI’s JSON Schema format: 
{
  "type": "function",
  "function": {
    "name": "calculate",
    "description": "Performs basic arithmetic operations.",
    "parameters": {
      "type": "object",
      "properties": {
        "operation": {"type": "string"},
        "a": {"type": "number"},
        "b": {"type": "number"}
      }
    }
  }
}

​
OAuth for MCP Servers

Some MCP servers require OAuth authentication. Configure OAuth tokens: 
OAUTH_TOKENS_PATH=.mcp_tokens
OAUTH_REDIRECT_URI=http://localhost:3000/callback
Tokens are stored in the OAUTH_TOKENS_PATH directory and automatically refreshed.

​
Troubleshooting

​
MCP Server Not Connecting

  1. Check server configuration in mcp_config.json
  2. Verify transport type matches server capabilities
  3. Check network connectivity for remote servers
  4. Review server logs for error messages

​
Tools Not Available

  1. Verify MCP server is loaded (check startup logs)
  2. Check allowed_tools includes the tool name
  3. Verify tool name matches exactly (case-sensitive)

​
Stdio Server Hanging

  • Ensure Node.js is installed: node --version
  • Check npx is available: npx --version
  • Verify command and args are correct

​
Remote Server Authentication

  • Check headers configuration in mcp_config.json
  • Verify token is valid and not expired
  • Ensure server supports the transport type

​
Best Practices

  1. Use local servers for development/testing
  2. Use remote servers for production (better reliability)
  3. Filter tools per request for security
  4. Monitor tool usage for rate limits
  5. Cache tool results when possible

​
Example: Full MCP Setup

{
  "mcpServers": {
    "Context7": {
      "url": "https://mcp.context7.com/mcp",
      "transport": "streamable-http"
    },
    "memory": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-memory"]
    },
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "."]
    }
  }
}
See MCP Servers Configuration for more details.