Research MCP Server

Use This MCP server To

Aggregate search results from multiple APIs into a single MCP context Enable LLMs to perform deep research using combined search engine data Provide real-time search context for AI-powered research assistants Deploy a scalable MCP server for multi-source research queries Integrate Brave and Tavily search APIs for enriched data retrieval Support development of AI agents with enhanced external knowledge access

README

ResearchMCP

A research tool that combines the Model Context Protocol (MCP) with Brave Search, Tavily Search, etc. to achieve research capabilities equivalent to (or aiming to be equivalent to) ChatGPT's DeepResearch.

Technology Stack

Runtime: Deno
Framework: Hono
Error Handling: Neverthrow (Result<T, E> pattern)
Deployment: Docker container

Setup

Local Development

Install Deno
Clone the repository
Set environment variables
- BRAVE_API_KEY: Brave Search API key
- TAVILY_API_KEY: Tavily Search API key (optional)

Run the application:

make dev  # HTTP server
make mcp  # MCP server for Claude Desktop

Container Environment

Install Docker and Docker Compose
Set environment variables in a .env file
Build and run the container:
```
make d-build
make d-up
```

Development Commands

# Local development
make dev     # Start HTTP server with watch mode
make mcp     # Start MCP server for Claude Desktop
make test    # Run tests
make lint    # Run linter
make format  # Format code
make check   # Type check

# Docker development
make d-build  # Build the image
make d-dev    # Start container with live reload
make d-up     # Run in background
make d-logs   # View logs
make d-down   # Stop container

MCP Integration with Claude Desktop

Run the MCP server: make mcp

In Claude Desktop, add a new MCP server with the following configuration:

{
  "mcpServers": {
    "MCPSearch": {
      "description": "Web search powered by Brave, Tavily, etc.",
      "command": "/absolute/path/to/ResearchMCP/cli.ts",
      "args": [],
      "transport": "stdio",
      "env": {
        "BRAVE_API_KEY": "your_brave_api_key_here",
        "TAVILY_API_KEY": "your_tavily_api_key_here"
      }
    }
  }
}

Replace /absolute/path/to/ResearchMCP/cli.ts with the actual path to the cli.ts file.

Known Limitations

Language Support: Brave Search API has limited support for non-Latin characters. Searches in Japanese, Chinese, Korean, and other non-Latin script languages may fail with encoding errors. For best results, use English queries.

Features

Web Search: Search the web using Brave Search API through Claude Desktop
MCP Protocol: Full compliance with the Model Context Protocol
Caching: Search results are cached to improve performance and reduce API calls

ResearchMCP FAQ

How do I set up ResearchMCP for local development?

Install Deno, clone the repo, set environment variables for API keys, then run 'make dev' or 'make mcp'.

Can ResearchMCP be deployed in a containerized environment?

Yes, it supports Docker and Docker Compose for easy container deployment.

What search APIs does ResearchMCP support?

It supports Brave Search and optionally Tavily Search, with potential for adding more.

How does ResearchMCP handle errors?

It uses the Neverthrow pattern (Result<T, E>) for robust error handling.

Is ResearchMCP compatible with multiple LLM providers?

Yes, it integrates with MCP clients that can use models like OpenAI, Claude, and Gemini.

What programming languages and frameworks is ResearchMCP built with?

It is built using Deno runtime and the Hono web framework.

How do I provide API keys to ResearchMCP?

Set environment variables such as BRAVE_API_KEY and optionally TAVILY_API_KEY before running the server.

Can ResearchMCP be used to enhance AI research assistants?

Yes, it aggregates multiple search sources to provide richer context for AI models.