web-research-assistant

The Web Research Assistant MCP server provides a suite of tools designed to help you search, crawl, and analyze the web, powered by a local SearXNG instance, an Exa neural search integration, and crawl4ai. It includes 13 tools for tasks like federated web search, code and package discovery, crawling, data extraction, and health checks, as well as 4 data lookup resources and 5 reusable prompts to streamline research workflows. You can access these tools through your MCP host (e.g., Claude Desktop, OpenCode) and wire them into your agents to perform complex web-research tasks with structured results and reliable error handling.

To use the server, install the MCP client and run the server component via uvx (Python) or install from PyPI. Once running, the MCP host can invoke tools such as web_search for federated queries across SearXNG, extract_data for structured extraction from pages, package_info and package_search for package metadata, translate_error for troubleshooting error messages, api_docs for automatic API documentation crawling, and check_service_status for multi-service health checks. The setup assumes you have a running SearXNG instance (recommended on http://localhost:2288) and optional integrations like Exa API keys or Pixabay keys for enhanced results.

Prerequisites\n- Python 3.10+\n- A running SearXNG instance (recommended on http://localhost:2288)\n- Optional: Exa API key for neural search and Pixabay API key for image search\n\nInstallation options\nOption A: Using uvx (recommended; no global installation required)\n1. Install and run the MCP server directly from PyPI:\nbash\nuvx web-research-assistant\n\n2. The server will start and listen for MCP client connections on stdio/stdin.\n\nOption B: Install with pip (global install)\n1. Install the package globally:\nbash\ pip install web-research-assistant\n\n2. Run the server:\nbash\nweb-research-assistant\n\n\nOption C: Install with uv (development or local usage)\n1. Install UV tooling if needed and install the tool:\nbash\nuv tool install web-research-assistant\n\n2. Run the server:\nbash\nweb-research-assistant\n\n\nClient configuration examples (MCP host integration)\nClaude Desktop (uvx recommended)\njson\n{\n "mcpServers": {\n "web-research-assistant": {\n "command": "uvx",\n "args": ["web-research-assistant"]\n }\n }\n}\n OpenCode (installed package)\njson\n{\n "mcp": {\n "web-research-assistant": {\n "type": "local",\n "command": ["web-research-assistant"],\n "enabled": true\n }\n }\n}\n

Environment and configuration tips:\n- Ensure your SearXNG instance is reachable at the expected URL (default http://localhost:2288).\n- For best results, enable search engines in SearXNG that align with your research needs (GitHub, Stack Overflow, code-focused engines, etc.).\n- If using Exa neural search, set up an API key for better results.\n- The server is designed to run via stdio, making it easy to wire into Claude Desktop or other MCP hosts. If deploying with Docker, follow the DOCKER_SETUP.md and ensure network access to SearXNG.\n- When configuring clients, you can choose to run web-research-assistant directly from the installed package or via uvx for development workflows.\n- Optional: Install Playwright browsers for advanced crawling with crawl4ai.\n- If you upgrade from an earlier version, review the changelogs to understand breaking changes or new prompts and tools.