MCP 服务器

Strands MCP Server

Bidirectional MCP integration that allows Strands agents to both expose themselves as MCP servers and connect to other MCP servers as clients, enabling seamless agent-to-agent communication and tool sharing.

README

strands-mcp-server

Bidirectional MCP integration for Strands Agents.

pip install strands-mcp-server

Overview

mcp_server - Expose agent as MCP server
mcp_client - Connect to MCP servers
CLI - stdio for Claude Desktop

graph LR
    subgraph "Your Strands Agent"
        A[Tools: calculator, shell, etc.]
        B[mcp_server tool]
        C[mcp_client tool]
        A --> B
        C --> A
    end
    
    subgraph "Server Mode"
        B -->|HTTP/stdio| D[MCP Protocol]
        D --> E[Claude Desktop]
        D --> F[Other Agents]
        D --> G[Custom Clients]
    end
    
    subgraph "Client Mode"
        H[Remote MCP Servers] -->|HTTP/stdio/SSE| I[MCP Protocol]
        I --> C
    end
    
    subgraph "CLI"
        J[uvx strands-mcp-server] -->|Local Mode| D
        J -->|Proxy Mode| I
    end
    
    style A fill:#2d3748,stroke:#4a5568,color:#fff
    style B fill:#2b6cb0,stroke:#2c5282,color:#fff
    style C fill:#38a169,stroke:#2f855a,color:#fff
    style D fill:#805ad5,stroke:#6b46c1,color:#fff
    style I fill:#805ad5,stroke:#6b46c1,color:#fff
    style E fill:#d69e2e,stroke:#b7791f,color:#fff
    style F fill:#d69e2e,stroke:#b7791f,color:#fff
    style G fill:#d69e2e,stroke:#b7791f,color:#fff
    style H fill:#e53e3e,stroke:#c53030,color:#fff
    style J fill:#48bb78,stroke:#38a169,color:#fff

Quick Start

Server:

from strands import Agent
from strands_mcp_server import mcp_server

agent = Agent(tools=[..., mcp_server])
agent("start mcp server on port 8000")

Client:

from strands import Agent
from strands_mcp_server import mcp_client

agent = Agent(tools=[mcp_client])
agent.tool.mcp_client(
    action="connect",
    connection_id="remote",
    transport="http",
    server_url="http://localhost:8000/mcp"
)
agent.tool.mcp_client(
    action="call_tool",
    connection_id="remote",
    tool_name="calculator",
    tool_args={"expression": "42 * 89"}
)

For Agents like Claude Desktop/Kiro/...:

{
  "mcpServers": {
    "my-agent": {
      "command": "uvx",
      "args": ["strands-mcp-server", "--cwd", "/path/to/project"]
    }
  }
}

API

mcp_server

Parameter	Default	Description
`action`	required	`start`, `stop`, `status`, `list`
`transport`	`http`	`http` or `stdio`
`port`	8000	Port
`tools`	None	Tools to expose (None = all)
`expose_agent`	True	Include `invoke_agent`
`stateless`	False	Multi-node ready

mcp_client

Parameter	Description
`action`	`connect`, `disconnect`, `list_tools`, `call_tool`
`connection_id`	Connection ID
`transport`	`http`, `stdio`, `sse`
`server_url`	Server URL
`tool_name`	Tool to call
`tool_args`	Tool arguments

invoke_agent

Full agent access when expose_agent=True:

agent.tool.mcp_client(
    action="call_tool",
    connection_id="remote",
    tool_name="invoke_agent",
    tool_args={"prompt": "Calculate 2 + 2"}
)

CLI

uvx strands-mcp-server [OPTIONS]

Option	Description
`--cwd PATH`	Working directory
`--upstream-url URL`	Upstream server (proxy)
`--system-prompt TEXT`	System prompt
`--no-agent-invocation`	Disable invoke_agent
`--debug`	Debug mode

Examples:

# Local
uvx strands-mcp-server --cwd /path/to/project

# Proxy
uvx strands-mcp-server --upstream-url http://localhost:8000/mcp

Troubleshooting

# Debug
uvx strands-mcp-server --cwd /path --debug

# Check connection
curl http://localhost:8000/mcp

# Port in use
lsof -i :8000 && kill -9 <PID>

# Claude logs
tail -f ~/Library/Logs/Claude/mcp*.log