Create venv

uv sync

Install package

uv pip install brave-mcp-langchain

Run MCP server in STDIO mode

uvx brave-mcp-langchain

To run MCP server in SSE mode

uvx brave-mcp-langchain sse 5003

MCP Setting

{
  "mcpServers": {
    "brave-mcp-langchain": {
      "disabled": false,
      "timeout": 60,
      "type": "stdio",
      "command": "uvx",
      "args": [
        "brave-mcp-langchain"
      ]
    }
  }
}

Use as Langchain tool

It can also be used as Langchain tool. Below is how to validate tool.

import httpx
import asyncio
from langchain.tools import Tool
from brave_mcp_langchain import brave_tool

async def test_search():
    result = await brave_tool.search_tool.ainvoke({"query": "LangGraph overview", "max_results": 10})
    print(result)

    result = await brave_tool.fetch_content_tool.ainvoke({
        "url": "https://iamatulsingh.github.io"
    })
    print(result)

asyncio.run(test_search())

Use with langchain example

import asyncio
from langchain.agents import initialize_agent
from langchain.agents.agent_types import AgentType
from langchain_ollama import ChatOllama
from brave_mcp_langchain import brave_tool

llm = ChatOllama(model="llama3.1:8b")

tools = [
    brave_tool.search_tool,
    brave_tool.fetch_content_tool
]

agent = initialize_agent(
    tools=[brave_tool.search_tool, brave_tool.fetch_content_tool],
    llm=llm,
    agent=AgentType.STRUCTURED_CHAT_ZERO_SHOT_REACT_DESCRIPTION,
    verbose=True
)

async def run_agent_query():
    response = await agent.ainvoke(
        "Search for 'iamatulsingh' overview, then fetch content from https://iamatulsingh.github.io"
    )
    print("\nAgent Response:")
    print(response)

asyncio.run(run_agent_query())

🧠 Inspiration & Attribution

This project, brave-mcp-langchain, was inspired by and partially based on the excellent work in duckduckgo-mcp-server by @nickclyde. That project laid the groundwork for integrating DuckDuckGo search and content fetching into the MCP ecosystem.

While brave-mcp-langchain extends the concept to support Brave Search and LangChain workflows, several architectural ideas and implementation patterns were adapted from duckduckgo-mcp-server, which is licensed under the MIT License.

I'm grateful for the open-source community and contributors who make projects like this possible. If you’re interested in DuckDuckGo-based search tools, definitely check out the original repository!