Implementation:CrewAIInc CrewAI MCP Server Config

Overview

Concrete configuration classes for connecting to MCP tool servers via stdio, HTTP, or SSE transport provided by the CrewAI framework.

Source Reference

Repository: crewAIInc/crewAI
File: lib/crewai/src/crewai/mcp/config.py
Lines: L12-124

Signatures

MCPServerStdio

class MCPServerStdio(BaseModel):
    command: str
    args: list[str] = []
    env: dict[str, str] | None = None
    tool_filter: ToolFilter | None = None
    cache_tools_list: bool = False

MCPServerHTTP

class MCPServerHTTP(BaseModel):
    url: str
    headers: dict[str, str] | None = None
    streamable: bool = True
    tool_filter: ToolFilter | None = None

MCPServerSSE

class MCPServerSSE(BaseModel):
    url: str
    headers: dict[str, str] | None = None
    tool_filter: ToolFilter | None = None

MCPServerConfig Type Alias

MCPServerConfig = MCPServerStdio | MCPServerHTTP | MCPServerSSE

Fields

MCPServerStdio Fields

Field	Type	Default	Description
`command`	`str`	(required)	The command to launch the MCP server process (e.g., `"npx"`, `"python"`)
`args`	`list[str]`	`[]`	Command-line arguments passed to the server process
`env`	None	`None`	Environment variables to set for the server process
`tool_filter`	None	`None`	Optional filter to restrict which tools are exposed from this server
`cache_tools_list`	`bool`	`False`	If `True`, cache the server's tool list to avoid re-querying on each connection

MCPServerHTTP Fields

Field	Type	Default	Description
`url`	`str`	(required)	The HTTP URL of the MCP server endpoint
`headers`	None	`None`	Optional HTTP headers (e.g., for authentication)
`streamable`	`bool`	`True`	Whether to use streamable HTTP transport
`tool_filter`	None	`None`	Optional filter to restrict which tools are exposed from this server

MCPServerSSE Fields

Field	Type	Default	Description
`url`	`str`	(required)	The SSE endpoint URL of the MCP server
`headers`	None	`None`	Optional HTTP headers (e.g., for authentication)
`tool_filter`	None	`None`	Optional filter to restrict which tools are exposed from this server

Import

from crewai.mcp import MCPServerStdio, MCPServerHTTP, MCPServerSSE

How It Works

Each configuration class captures the information needed to establish a connection to an MCP server using a specific transport:

MCPServerStdio launches a local process using the specified command and args. Communication occurs over the process's standard input/output streams. This is commonly used with npx to launch Node.js-based MCP servers.
MCPServerHTTP connects to a remote HTTP endpoint. The streamable flag controls whether bidirectional streaming is used. Authentication headers can be provided.
MCPServerSSE connects to an SSE endpoint for real-time streaming responses. This is useful for servers that push incremental updates during long-running tool operations.

The MCPServerConfig type alias is a union of all three, used in agent configuration to accept any transport type.

The optional tool_filter field accepts a ToolFilter callable or configuration that restricts which of the server's tools are made available to the agent. This is important when an MCP server exposes many tools but only a subset is relevant.

Example

MCPServerStdio with npx Command

from crewai import Agent
from crewai.mcp import MCPServerStdio

# Configure a filesystem MCP server launched via npx
filesystem_server = MCPServerStdio(
    command="npx",
    args=["-y", "@modelcontextprotocol/server-filesystem", "/home/user/documents"],
    env={"NODE_ENV": "production"},
    cache_tools_list=True,
)

agent = Agent(
    role="Document Analyst",
    goal="Read and analyze documents from the filesystem",
    backstory="A meticulous analyst who reviews documents carefully.",
    mcp_servers=[filesystem_server],
)

MCPServerHTTP with Remote URL

from crewai import Agent
from crewai.mcp import MCPServerHTTP

# Configure a remote MCP server with authentication
api_server = MCPServerHTTP(
    url="https://mcp.example.com/tools",
    headers={"Authorization": "Bearer your-api-token"},
    streamable=True,
)

agent = Agent(
    role="API Integrator",
    goal="Interact with external services via MCP tools",
    backstory="A skilled integrator who connects to external APIs.",
    mcp_servers=[api_server],
)

MCPServerSSE for Streaming

from crewai.mcp import MCPServerSSE

# Configure an SSE-based MCP server
sse_server = MCPServerSSE(
    url="https://mcp-stream.example.com/sse",
    headers={"X-API-Key": "your-key"},
)

Combining Multiple MCP Servers

from crewai import Agent
from crewai.mcp import MCPServerStdio, MCPServerHTTP

agent = Agent(
    role="Full-Stack Agent",
    goal="Use both local and remote tool servers",
    backstory="An agent with access to diverse tool sources.",
    mcp_servers=[
        MCPServerStdio(
            command="npx",
            args=["-y", "@modelcontextprotocol/server-filesystem", "/data"],
        ),
        MCPServerHTTP(
            url="https://mcp.example.com/tools",
            headers={"Authorization": "Bearer token"},
        ),
    ],
)

Principle Link

Principle:CrewAIInc_CrewAI_MCP_Server_Connection

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment