ReActMCP Web Search

ReActMCP Web Search is an MCP (Model Context Protocol) server that integrates web search capabilities into your AI assistant framework. It leverages the Exa API to perform both basic and advanced web searches, returning real-time, markdown-formatted results including titles, URLs, publication dates, and content summaries.

This repository is part of the broader ReActMCP project that connects various MCP tools and servers to empower your AI assistant with a wide range of capabilities.

- Features
- Requirements
- Installation
- Configuration

- Environment Variables
- MCP Configuration
- System Prompt

- Running the Web Search Server
- Testing the Tools

- Python 3.8+
- python-dotenv
- exa-py(Exa API client)
- firecrawl-py(Firecrawl API client)
- Other dependencies that may be required by your MCP framework

git clone https://github.com/mshojaei77/ReActMCP.git cd ReActMCP

python -m venv venv source venv/bin/activate # On Windows use: venv\Scripts\activate

EXA_API_KEY=your_exa_api_key_here OPENAI_API_KEY=...

{ "websearch": { "script": "web_search.py", "encoding_error_handler": "ignore", "description": "Web search capability using Exa API that provides real-time internet search results. Supports both basic and advanced search with filtering options including domain restrictions, text inclusion requirements, and date filtering. Returns formatted results with titles, URLs, publication dates, and content summaries.", "required_env_vars": ["EXA_API_KEY"], "active": true }, "settings": { "model": "gpt-4o", "system_prompt_path": "system_prompt.txt" } }

You can personalize or extend this configuration by modifying parameters such as default number of results or adding new MCP tools.

Thesystem_prompt.txtfile configures the behavior and tone of your AI assistant. It guides responses to be friendly, engaging, and informative, with the inclusion of emojis. An example prompt is provided:

You are a helpful, knowledgeable AI assistant with web search capabilities. Your goal is to provide accurate, comprehensive, and up-to-date information to users. Use lots of emojis and make your responses fun and engaging. ## Available Search Tools - search_web: Basic web search that returns results based on a query - advanced_search_web: Advanced search with filtering options for domains, required text, and date ranges ## Guidelines for Responding to Questions 1. For current information or facts that might have changed since your training data, use the appropriate search tool to find the most recent and relevant information. 2. Use search_web for general queries and advanced_search_web with appropriate filters for more specific needs. 3. Formulate precise search queries to maximize result relevance. 4. For recent information, use the max_age_days parameter in advanced search to limit results to recent publications. 5. When targeting specific sources, use the include_domains parameter to focus your search. 6. Cite sources by including URLs from search results. 7. For insufficient or contradictory results, acknowledge limitations and explain findings. 8. Break down complex topics into organized sections. 9. Provide balanced perspectives on controversial topics. 10. Be transparent about uncertainty rather than making up information. 11. Maintain a helpful, informative, and conversational tone. ## Response Quality Standards Your responses should be well-structured, factually accurate, and tailored to the user's level of understanding on the topic. Use the web search capabilities as your primary tools for accessing current information before responding to time-sensitive or factual queries.

Feel free to adjust the system prompt to align with your desired assistant behavior.

The published container runs the Exa MCP server over stdio. Build and run it with the API key supplied at runtime:

docker build -t reactmcp-exa:local . docker run --rm -i -e EXA_API_KEY=your_exa_api_key_here reactmcp-exa:local

The exact container command for MCP hosts is:

docker run --rm -i -e EXA_API_KEY=your_exa_api_key_here reactmcp-exa:local

The Docker image entry point isservers/exa_web_search.py.main.pyis the optional OpenAI/LangGraph host and is not the container MCP entry point.

The repository includesglama.jsonwith the maintainer identity used by Glama's claim flow. After merging it, repeat the Glama claim flow so Glama syncs the metadata and Dockerfile.

To validate the image with MCP Inspector:

npx @modelcontextprotocol/inspector docker run --rm -i -e EXA_API_KEY=your_exa_api_key_here reactmcp-exa:local

Run the Inspector command from a terminal with Docker and Node.js installed.

The MCP servers is implemented inserversdirectory. To run a server, simply execute it :

This command will start the MCP server which listens for requests and exposes the following tools:

- search_web: Perform basic web searches.
- advanced_search_web: Perform advanced web searches with filtering options.

Withinservers/exa_web_search.py, a test functiontest_search()is provided (currently commented out) to demonstrate basic usage of the search capabilities. You can run this test by uncommenting the test execution block and using Python's asyncio runner:

if __name__ == "__main__": import asyncio # Uncomment the following line to perform a test search # asyncio.run(test_search()) mcp.run()

This will print search results for sample queries and help you verify that the tool is functioning as expected.

Configure Claude Desktop to use this server by adding the following to your configuration:

{ "mcpServers": { "websearch": { "command": "python", "args": ["path/to/servers/exa_web_search.py"] } } }

- Missing EXA_API_KEY:Ensure that the.envfile is properly set up with your valid Exa API key.
- Dependency Issues:Verify that all necessary Python packages are installed (check yourrequirements.txtfile). Reinstall packages if needed.
- API Errors:If you encounter errors during web searches, check your network connection and verify the Exa API status.

This project is licensed under the MIT License. See theLICENSEfile for details.

SerpApi MCP Server for Google and other search engine results