​
Adapters API Reference

Adapters provide a bridge between MCP tools and different agent frameworks. They convert MCP tool definitions into framework-specific formats.

​
BaseAdapter

The abstract base class for all adapters.

​
BaseAdapter()

Base adapter class that defines the interface for tool conversion. Methods:

​
async create_tools(client: MCPClient, **kwargs) -> List[Any]

Converts MCP tools to framework-specific tool format. Parameters:
  • client (MCPClient): The MCP client instance
  • **kwargs: Additional framework-specific parameters
Returns:
  • List[Any]: List of framework-specific tool objects
Raises:
  • NotImplementedError: Must be implemented by subclasses

​
LangChainAdapter

Adapter for LangChain framework integration.

​
LangChainAdapter()

Creates tools compatible with LangChain agents and runnables. Example: 
from mcp_use import MCPClient
from mcp_use.adapters import LangChainAdapter

client = MCPClient.from_config_file("config.json")
adapter = LangChainAdapter()
tools = await adapter.create_tools(client)

​
Methods

​
async create_tools(client: MCPClient, allowed_tools=None, disallowed_tools=None) -> List[BaseTool]

Converts MCP tools to LangChain BaseTool objects. Parameters:
  • client (MCPClient): The MCP client instance
  • allowed_tools (List[str], optional): Whitelist of tool names to include
  • disallowed_tools (List[str], optional): Blacklist of tool names to exclude
Returns:
  • List[BaseTool]: List of LangChain tool objects
Example: 
# Create all available tools
tools = await adapter.create_tools(client)

# Create only specific tools
tools = await adapter.create_tools(
    client,
    allowed_tools=["file_read", "file_write"]
)

# Exclude dangerous tools
tools = await adapter.create_tools(
    client,
    disallowed_tools=["system_execute", "file_delete"]
)

​
async create_tool(client: MCPClient, tool_definition: dict) -> BaseTool

Creates a single LangChain tool from an MCP tool definition. Parameters:
  • client (MCPClient): The MCP client instance
  • tool_definition (dict): MCP tool definition
Returns:
  • BaseTool: LangChain tool object
Example: 
tool_def = {
    "name": "file_read",
    "description": "Read file contents",
    "inputSchema": {
        "type": "object",
        "properties": {
            "path": {"type": "string"}
        },
        "required": ["path"]
    }
}

tool = await adapter.create_tool(client, tool_def)

​
Tool Filtering

Adapters support various filtering mechanisms to control which tools are available.

​
Whitelist Filtering

Only include specified tools: 
adapter = LangChainAdapter()
tools = await adapter.create_tools(
    client,
    allowed_tools=[
        "file_read",
        "file_write",
        "web_search",
        "sqlite_query"
    ]
)

​
Blacklist Filtering

Exclude potentially dangerous tools: 
adapter = LangChainAdapter()
tools = await adapter.create_tools(
    client,
    disallowed_tools=[
        "system_execute",
        "file_delete",
        "network_request",
        "process_kill"
    ]
)

​
Pattern-Based Filtering

Use patterns for flexible filtering: 
import re
from typing import Any
from mcp.types import Tool
from ..connectors.base import BaseConnector

class PatternLangChainAdapter(LangChainAdapter):
    def __init__(self, allowed_patterns=None, disallowed_patterns=None):
        super().__init__()
        self.allowed_patterns = allowed_patterns or []
        self.disallowed_patterns = disallowed_patterns or []

    def _convert_tool(self, mcp_tool: Tool, connector: BaseConnector) -> Any:
        """Override to add pattern-based filtering."""
        tool_name = mcp_tool.name

        # Check allowed patterns
        if self.allowed_patterns:
            if not any(re.match(pattern, tool_name) for pattern in self.allowed_patterns):
                return None

        # Check disallowed patterns
        if self.disallowed_patterns:
            if any(re.match(pattern, tool_name) for pattern in self.disallowed_patterns):
                return None

        # Use parent class conversion
        return super()._convert_tool(mcp_tool, connector)

# Usage
adapter = PatternLangChainAdapter(
    allowed_patterns=[r"file_.*", r"sqlite_.*"],  # Only file and sqlite tools
    disallowed_patterns=[r".*_delete", r".*_execute"]  # No delete or execute tools
)
tools = await PatternLangChainAdapter.create_tools(client)

​
Custom Adapters

Create custom adapters for other frameworks:

​
CrewAI Adapter Example

from mcp_use.adapters import BaseAdapter

class CrewAIAdapter(BaseAdapter):
    def _convert_tool(self, mcp_tool: Tool, connector: BaseConnector):
        """Convert MCP tool to CrewAI format."""
        from crewai_tools import BaseTool as CrewAIBaseTool

        class MCPCrewAITool(CrewAIBaseTool):
            name = mcp_tool.name
            description = mcp_tool.description

            async def _run(self, **kwargs):
                return await connector.call_tool(self.name, kwargs)

        return MCPCrewAITool()

    def _convert_resource(self, mcp_resource: Resource, connector: BaseConnector):
        """Convert MCP resource to CrewAI format - implement as needed."""
        return None  # Skip resources for this example

    def _convert_prompt(self, mcp_prompt: Prompt, connector: BaseConnector):
        """Convert MCP prompt to CrewAI format - implement as needed."""
        return None  # Skip prompts for this example

# Usage
tools = await CrewAIAdapter.create_tools(client)

​
AutoGen Adapter Example

class AutoGenAdapter(BaseAdapter):
    def _convert_tool(self, mcp_tool: Tool, connector: BaseConnector):
        """Convert MCP tool to AutoGen format."""
        async def implementation(**kwargs):
            return await connector.call_tool(mcp_tool.name, kwargs)

        return {
            "type": "function",
            "function": {
                "name": mcp_tool.name,
                "description": mcp_tool.description,
                "parameters": mcp_tool.inputSchema or {},
                "implementation": implementation
            }
        }

    def _convert_resource(self, mcp_resource: Resource, connector: BaseConnector):
        """Convert MCP resource to AutoGen format - implement as needed."""
        return None  # Skip resources for this example

    def _convert_prompt(self, mcp_prompt: Prompt, connector: BaseConnector):
        """Convert MCP prompt to AutoGen format - implement as needed."""
        return None  # Skip prompts for this example

# Usage
tools = await AutoGenAdapter.create_tools(client)

​
Tool Metadata

Adapters preserve and enhance tool metadata:

​
Accessing Tool Information

adapter = LangChainAdapter()
tools = await adapter.create_tools(client)

for tool in tools:
    print(f"Name: {tool.name}")
    print(f"Description: {tool.description}")
    print(f"Args Schema: {tool.args}")

    # Access MCP-specific metadata
    if hasattr(tool, '_mcp_server'):
        print(f"Server: {tool._mcp_server}")
    if hasattr(tool, '_mcp_original_schema'):
        print(f"Original Schema: {tool._mcp_original_schema}")

​
Enhanced Tool Descriptions

class EnhancedLangChainAdapter(LangChainAdapter):
    async def create_tool(self, client, tool_def):
        tool = await super().create_tool(client, tool_def)

        # Enhance description with usage examples
        if "examples" in tool_def:
            enhanced_description = f"{tool.description}\n\nExamples:\n"
            for example in tool_def["examples"]:
                enhanced_description += f"- {example}\n"
            tool.description = enhanced_description

        # Add safety warnings
        if self._is_dangerous_tool(tool_def["name"]):
            tool.description += "\n⚠️ WARNING: This tool performs potentially dangerous operations."

        return tool

    def _is_dangerous_tool(self, tool_name):
        dangerous_patterns = ["delete", "execute", "kill", "remove", "destroy"]
        return any(pattern in tool_name.lower() for pattern in dangerous_patterns)

​
Error Handling

Adapters handle various error conditions:

​
Tool Creation Errors

class RobustLangChainAdapter(LangChainAdapter):
    def _convert_tool(self, mcp_tool: Tool, connector: BaseConnector):
        """Override with error handling for individual tool conversion."""
        try:
            return super()._convert_tool(mcp_tool, connector)
        except Exception as e:
            logger.error(f"Failed to create tool {mcp_tool.name}: {e}")
            return None  # Skip this tool

    async def load_tools_for_connector(self, connector: BaseConnector):
        """Override to add robust error handling."""
        try:
            return await super().load_tools_for_connector(connector)
        except Exception as e:
            logger.error(f"Failed to load tools for connector: {e}")
            return []  # Return empty list on failure

​
Runtime Error Handling

class SafeLangChainAdapter(LangChainAdapter):
    def _convert_tool(self, mcp_tool: Tool, connector: BaseConnector):
        """Override to create tools with enhanced error handling."""
        if mcp_tool.name in self.disallowed_tools:
            return None

        adapter_self = self

        class SafeMcpTool(BaseTool):
            name: str = mcp_tool.name
            description: str = mcp_tool.description
            tool_connector: BaseConnector = connector
            handle_tool_error: bool = True

            def _run(self, **kwargs):
                raise NotImplementedError("MCP tools only support async operations")

            async def _arun(self, **kwargs):
                try:
                    result = await self.tool_connector.call_tool(self.name, kwargs)
                    return adapter_self._parse_mcp_tool_result(result)
                except ToolExecutionError as e:
                    return f"Tool execution failed: {e}"
                except TimeoutError:
                    return "Tool execution timed out"
                except Exception as e:
                    return f"Unexpected error: {e}"

        return SafeMcpTool()

​
Performance Optimization

​
Lazy Tool Creation

class LazyLangChainAdapter(LangChainAdapter):
    def __init__(self, disallowed_tools=None):
        super().__init__(disallowed_tools)
        self._tool_cache = {}

    def _convert_tool(self, mcp_tool: Tool, connector: BaseConnector):
        """Override to create lazy-loading tools."""
        if mcp_tool.name in self.disallowed_tools:
            return None

        adapter_self = self

        class LazyMcpTool(BaseTool):
            name: str = mcp_tool.name
            description: str = mcp_tool.description
            tool_connector: BaseConnector = connector

            def _run(self, **kwargs):
                raise NotImplementedError("MCP tools only support async operations")

            async def _arun(self, **kwargs):
                # Create actual tool on first use
                if self.name not in adapter_self._tool_cache:
                    adapter_self._tool_cache[self.name] = await self._create_actual_tool()

                return await adapter_self._tool_cache[self.name].execute(**kwargs)

            async def _create_actual_tool(self):
                # Create the actual tool implementation
                result = await self.tool_connector.call_tool(self.name, {})
                return adapter_self._parse_mcp_tool_result(result)

        return LazyMcpTool()

​
Batch Tool Creation

class BatchLangChainAdapter(LangChainAdapter):
    def __init__(self, batch_size=10, disallowed_tools=None):
        super().__init__(disallowed_tools)
        self.batch_size = batch_size

    async def load_tools_for_connector(self, connector: BaseConnector):
        """Override to process tools in batches."""
        # Check if we already have tools for this connector
        if connector in self._connector_tool_map:
            return self._connector_tool_map[connector]

        # Ensure connector is initialized
        success = await self._ensure_connector_initialized(connector)
        if not success:
            return []

        connector_tools = []
        all_tools = connector.tools or []

        # Process tools in batches
        for i in range(0, len(all_tools), self.batch_size):
            batch = all_tools[i:i + self.batch_size]
            batch_tools = []

            for tool in batch:
                converted_tool = self._convert_tool(tool, connector)
                if converted_tool:
                    batch_tools.append(converted_tool)

            connector_tools.extend(batch_tools)

        # Store the tools for this connector
        self._connector_tool_map[connector] = connector_tools
        return connector_tools

​
Integration Examples

​
LangChain Agent

from langchain.agents import create_tool_calling_agent, AgentExecutor
from langchain.prompts import ChatPromptTemplate
from langchain_openai import ChatOpenAI

async def create_langchain_agent():
    # Create MCP client and adapter
    client = MCPClient.from_config_file("config.json")
    tools = await LangChainAdapter.create_tools(client)

    # Create LLM and prompt
    llm = ChatOpenAI(model="gpt-4")
    prompt = ChatPromptTemplate.from_messages([
        ("system", "You are a helpful assistant with access to tools."),
        ("human", "{input}"),
        ("placeholder", "{agent_scratchpad}"),
    ])

    # Create agent
    agent = create_tool_calling_agent(llm, tools, prompt)
    agent_executor = AgentExecutor(agent=agent, tools=tools)

    return agent_executor

# Usage
agent = await create_langchain_agent()
result = await agent.ainvoke({"input": "Read the contents of README.md"})

​
Custom Framework Integration

class CustomFrameworkTool:
    def __init__(self, name, description, execute_fn):
        self.name = name
        self.description = description
        self.execute_fn = execute_fn

    async def execute(self, **kwargs):
        return await self.execute_fn(**kwargs)

class CustomFrameworkAdapter(BaseAdapter):
    def _convert_tool(self, mcp_tool: Tool, connector: BaseConnector):
        """Convert MCP tool to custom framework format."""
        async def executor(**kwargs):
            return await connector.call_tool(mcp_tool.name, kwargs)

        return CustomFrameworkTool(
            name=mcp_tool.name,
            description=mcp_tool.description,
            execute_fn=executor
        )

    def _convert_resource(self, mcp_resource: Resource, connector: BaseConnector):
        """Convert MCP resource to custom framework format."""
        return None  # Skip resources for this example

    def _convert_prompt(self, mcp_prompt: Prompt, connector: BaseConnector):
        """Convert MCP prompt to custom framework format."""
        return None  # Skip prompts for this example

# Usage
tools = await CustomFrameworkAdapter.create_tools(client)

for tool in tools:
    result = await tool.execute(param1="value1")

​
See Also