Brave Deep Research MCP
by suthio
Ask Claude to research a topic and get full page content, not just snippets. Brave Deep Research MCP runs a Brave Search, then uses Puppeteer to visit each result, extract the main content, and optionally follow links to a configurable depth. You get back cleaned text from the actual pages — no more "here's a 200-character snippet, good luck." When you need more than search-result blurbs — due diligence, competitive research, or compiling sources — this server turns Claude into a researcher that actually reads the pages. It strips nav, ads, and footers and returns structured content with titles and URLs. You control how many results to process (up to 10) and how many link levels to traverse (1–3). One tool call: "deep-search for the latest quantum computing breakthroughs with depth 2" and you get full-text from the top results plus one hop of linked pages. Setup: Brave Search API key (same as the standard Brave MCP), optional env for headless mode and timeouts. Install via npm global or npx @suthio/brave-deep-research-mcp. Best paired with Claude Desktop or Cursor; add the server to your MCP config with BRAVE_API_KEY. The tradeoff: Puppeteer means slower responses and more resource use than API-only search, but for deep research the payoff is real content instead of snippets.
Installation
Key Features
- ✓Deep search: extract full page content from Brave Search results via Puppeteer
- ✓Configurable depth: 1–3 levels of link traversal from each result
- ✓Main-content extraction: strips nav, ads, footers; returns clean text
- ✓Metadata: titles, descriptions, and URLs for each extracted page
- ✓Up to 10 results per query with configurable result count
- ✓Headless mode toggle and page timeout settings
Use Cases
- →Due diligence or competitive research requiring full articles, not snippets
- →Compiling multi-source reports from search results and their linked pages
- →Fact-checking or verification by reading full page content in context
- →Research workflows where search snippets are too shallow for decisions
- →Building datasets or summaries from live web content beyond API snippets