MCP Client Compatibility

Documentation Index

Fetch the complete documentation index at: https://docs.ifalabs.com/llms.txt

Use this file to discover all available pages before exploring further.

​

Compatibility Matrix

​

Claude Desktop

​

Configuration

{
  "mcpServers": {
    "ifalabs": {
      "command": "ifa-mcp",
      "args": ["--network", "base-mainnet"]
    }
  }
}

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json
• Linux: ~/.config/Claude/claude_desktop_config.json

​

Behaviour

​

Known Limitations

• Configuration requires a full restart. Changes to claude_desktop_config.json are not picked up until Claude Desktop is fully quit and relaunched. A settings reload is not sufficient.
• One active config file. Claude Desktop reads from a single config file. If you need to switch between mainnet and testnet configurations, update the --network argument in the config file and restart.
• No project-level config. Unlike Cursor, Claude Desktop does not support project-scoped MCP configuration. The config file is global.

​

Cursor

​

Configuration

• Name: ifalabs
• Type: command
• Command: ifa-mcp --network base-mainnet

{
  "mcpServers": {
    "ifalabs": {
      "command": "node",
      "args": ["C:/Users/YOUR_USERNAME/OneDrive/Desktop/ifamcpserver/dist/index.js", "--network", "base-mainnet"]
    }
  }
}

{
  "mcpServers": {
    "ifalabs": {
      "command": "npx",
      "args": ["-y", "@ifalabs/mcp-server", "--network", "base-mainnet"]
    }
  }
}

​

Behaviour

​

Known Limitations

• Windows binary resolution. On Windows, Cursor cannot resolve the global ifa-mcp binary because npm wraps global binaries in a .cmd file that Cursor does not execute. Use the node format with the full path to dist/index.js, or use npx -y @ifalabs/mcp-server instead. macOS and Linux users can use ifa-mcp directly.
• Composer context window. In long Composer sessions, the model may not trigger MCP tool calls if the context window is heavily loaded with code. Start a fresh Composer session if tool calls stop firing unexpectedly.
• Tool approval prompt. Depending on your Cursor settings, the first tool call in a session may require explicit approval before executing. This is a Cursor privacy setting, not an IFÁ Labs limitation.
• Project config requires Cursor restart. Changes to .cursor/mcp.json require a full Cursor restart to take effect.

​

Windsurf

​

Configuration

{
  "mcpServers": {
    "ifalabs": {
      "command": "ifa-mcp",
      "args": ["--network", "base-mainnet"]
    }
  }
}

​

Behaviour

​

Known Limitations

• Write mode tool calls are non-interactive. In Windsurf’s Write mode, tool results are used by the model directly without surfacing them to the user as intermediate steps. If you need to see raw tool output, use Chat mode.
• Server restart required after config change. Like other clients, Windsurf requires a full restart after MCP config changes.

​

TRAE

​

Configuration

{
  "mcpServers": {
    "ifalabs": {
      "command": "ifa-mcp",
      "args": ["--network", "base-mainnet"]
    }
  }
}

​

Behaviour

​

Known Limitations

• Tool result formatting. TRAE renders MCP tool results as plain text without markdown formatting. Structured output like price tables will display as unformatted text. This is a TRAE rendering limitation, not a server issue.
• Restart required after config changes.

​

Zed

​

Configuration

{
  "context_servers": {
    "ifalabs": {
      "command": {
        "path": "ifa-mcp",
        "args": ["--network", "base-mainnet"]
      }
    }
  }
}

​

Behaviour

​

Known Limitations

• Partial streaming support. Zed’s context server implementation streams text responses but may not stream intermediate tool call results progressively. Full results appear once the tool call completes.
• Zed version requirement. MCP context server support requires Zed 0.150 or higher. Earlier versions do not support the MCP stdio transport.
• No project-level config. Zed MCP configuration is global, set in the user-level settings.json.

​

Continue

​

Configuration

{
  "mcpServers": [
    {
      "name": "ifalabs",
      "command": "ifa-mcp",
      "args": ["--network", "base-mainnet"]
    }
  ]
}

​

Behaviour

​

Known Limitations

• IDE restart may be required. Depending on your IDE and Continue version, MCP config changes may require a full IDE restart rather than just a Continue reload.
• JetBrains support is newer. Continue’s MCP integration is more mature in VS Code. JetBrains users may encounter occasional tool call failures in complex multi-step workflows. If issues occur, retry the prompt.

​

LangChain

​

Configuration

import { Client } from "@modelcontextprotocol/sdk/client/index.js";
import { StdioClientTransport } from "@modelcontextprotocol/sdk/client/stdio.js";
import { ChatAnthropic } from "@langchain/anthropic";
import { tool } from "@langchain/core/tools";

// Connect to IFÁ Labs MCP Server
const transport = new StdioClientTransport({
  command: "ifa-mcp",
  args: ["--network", "base-mainnet"],
});

const mcpClient = new Client(
  { name: "langchain-agent", version: "1.0.0" },
  { capabilities: {} }
);

await mcpClient.connect(transport);

// Discover available tools
const { tools: mcpTools } = await mcpClient.listTools();

// Convert MCP tools to LangChain tool format
const langchainTools = mcpTools.map((mcpTool) =>
  tool(
    async (input) => {
      const result = await mcpClient.callTool({
        name: mcpTool.name,
        arguments: input,
      });
      return JSON.stringify(result.content);
    },
    {
      name: mcpTool.name,
      description: mcpTool.description,
      schema: mcpTool.inputSchema,
    }
  )
);

// Use with a LangChain agent
const model = new ChatAnthropic({ model: "claude-sonnet-4-20250514" });
const agentWithTools = model.bindTools(langchainTools);

​

Behaviour

​

Known Limitations

• Manual tool loop required. Unlike GUI clients, LangChain requires you to implement the tool execution loop explicitly. Tool calls are not automatic.
• Schema compatibility. LangChain’s tool schema format differs slightly from the raw MCP schema. The conversion pattern shown above handles this correctly — do not pass raw MCP tool definitions directly to LangChain without conversion.

​

CrewAI

​

Configuration

from mcp import ClientSession, StdioServerParameters
from mcp.client.stdio import stdio_client
from crewai_tools import tool
import asyncio
import json

server_params = StdioServerParameters(
    command="ifa-mcp",
    args=["--network", "base-mainnet"],
)

async def call_ifa_tool(tool_name: str, arguments: dict) -> dict:
    async with stdio_client(server_params) as (read, write):
        async with ClientSession(read, write) as session:
            await session.initialize()
            result = await session.call_tool(tool_name, arguments)
            return json.loads(result.content[0].text)

@tool("Get IFA Labs Asset Price")
def get_ifa_price(asset: str) -> str:
    """Fetch a live stablecoin price from IFÁ Labs oracle on Base Mainnet."""
    result = asyncio.run(call_ifa_tool("get_asset_price", {"asset": asset}))
    return f"{result['asset']}: ${result['price']} (updated {result['age_seconds']}s ago)"

​

Known Limitations

• Async bridging. CrewAI tools are synchronous by default. The asyncio.run pattern shown above bridges the async MCP client into a synchronous CrewAI tool. For production agents, consider using an async CrewAI setup to avoid event loop conflicts.
• One session per call. The example above opens a new MCP session for every tool call. For high-frequency agents, implement a persistent session with connection pooling.

​

AutoGen

​

Configuration

from autogen import AssistantAgent, UserProxyAgent
from mcp import ClientSession, StdioServerParameters
from mcp.client.stdio import stdio_client
import asyncio
import json

server_params = StdioServerParameters(
    command="ifa-mcp",
    args=["--network", "base-mainnet"],
)

async def get_ifa_price(asset: str, network: str = "base-mainnet") -> dict:
    async with stdio_client(server_params) as (read, write):
        async with ClientSession(read, write) as session:
            await session.initialize()
            result = await session.call_tool(
                "get_asset_price",
                {"asset": asset, "network": network}
            )
            return json.loads(result.content[0].text)

# Register as an AutoGen function tool
llm_config = {
    "functions": [
        {
            "name": "get_ifa_price",
            "description": "Fetch a live stablecoin price from the IFÁ Labs oracle.",
            "parameters": {
                "type": "object",
                "properties": {
                    "asset": {
                        "type": "string",
                        "description": "Asset symbol e.g. USDT/USD, CNGN/USD"
                    },
                    "network": {
                        "type": "string",
                        "description": "Target network. Defaults to base-mainnet."
                    }
                },
                "required": ["asset"]
            }
        }
    ]
}

​

Known Limitations

• Function registration is per-tool. AutoGen requires each MCP tool to be registered individually as a function. For full IFÁ Labs tool coverage, register all seven tools using the pattern above.
• Async session management. Same async bridging considerations as CrewAI apply here.

​

General Compatibility Notes

{
  "command": "ifa-mcp",
  "args": ["--network", "base-mainnet"]
}

{
  "command": "npx",
  "args": ["-y", "@ifalabs/mcp-server", "--network", "base-mainnet"]
}

​

Reporting Compatibility Issues

• GitHub: IFA-Labs/oracle_contract/issues — open an issue with your client name, version, operating system, and the error output from ifa-mcp --log-level debug
• Telegram: t.me/ifalabs
• Email: support@ifalabs.com

​

Next Steps