MCP Web Search Tool

What is MCP Web Search Tool?

MCP Web Search Tool is an MCP server that lets AI assistants perform live web searches, read full page content, and retrieve source citations. It uses stdio transport, has pluggable providers, and requires no scraper dependencies.

How to use MCP Web Search Tool?

Clone the repository, install dependencies with npm, copy the .env.example file and optionally set BRAVE_API_KEY, then build and start. Docker is also supported. For client integration (Claude Desktop, VS Code, Cursor, etc.), see MCP_CLIENTS.md.

Key features of MCP Web Search Tool

• Five tools: web_search, news_search, image_search, fetch_url, list_providers
• Search returns ranked summaries with stable IDs and structured JSON
• Primary provider is Brave Search; keyless DuckDuckGo fallback for web search
• fetch_url strips scripts, styles, nav, footer, and aside from page text
• Configurable caching (LRU+TTL) for both search and URL fetch results
• Environment‑driven configuration for timeouts, limits, and provider selection

Use cases of MCP Web Search Tool

• Get current, source‑backed answers to live questions during a conversation
• Research topics by searching and reading full articles with cited sources
• Find recent news with source name and publish date via news_search
• Retrieve images with thumbnails for visual queries
• Integrate real‑time web data into AI assistant workflows without additional scraping tools

FAQ from MCP Web Search Tool

What are the runtime requirements?

Node.js >= 20.18 and npm >= 10. The server uses native fetch and has no other scraper dependencies.

Do I need a Brave Search API key?

No. Without a key, web_search runs on DuckDuckGo as a keyless fallback. However, news_search and image_search require a Brave key. Set ALLOW_KEYLESS=false to refuse startup without a key.

What does fetch_url refuse to fetch?

It rejects non‑http(s) schemes and any host that resolves to private, loopback, link‑local, multicast, or IPv4‑mapped IPv6 private addresses (see SECURITY.md).

Which search providers are available?

Brave Search (default when BRAVE_API_KEY is set) and DuckDuckGo (keyless fallback). Call list_providers to check availability of news and image search in the current session.

How can I configure caching?

Via environment variables: CACHE_MAX_ENTRIES (default 256) and CACHE_TTL_MS (default 300 000) for search cache; FETCH_CACHE_MAX (default 128) and FETCH_CACHE_TTL_MS (default 600 000) for URL fetch cache.