Best web search & scraping MCP server for Cursor, tested

Dated results

• Open-WebSearch passed in Cursor on 2026-10-21 (Cursor CLI 2026.10.19-6bbf2c4, open-websearch 2.2.0, connected in 2.2s). Full result
• Fetch (reference) passed in Cursor on 2026-10-19 (Cursor CLI 2026.10.16-e3cb207, mcp-server-fetch 2026.10.19 (pinned in the command we ran), connect time not recorded). Full result
• DuckDuckGo search passed in Cursor on 2026-10-13 (Cursor CLI 2026.10.12-d47ca18, package version not recorded, connect time not recorded). Full result
• Reddit MCP Buddy partial in Cursor on 2026-10-15 (Cursor CLI 2026.10.12-d47ca18, server reported v1.1.14, connect time not recorded). Full result
• MarkItDown MCP (Microsoft) failed in Cursor on 2026-10-21 (Cursor CLI 2026.10.19-6bbf2c4, markitdown-mcp 0.0.1a8, connected in 2.0s). Full result

Why Open-WebSearch is the candidate

Open-WebSearch passed every step in Cursor. It is the group’s site-wide candidate (judged 2026-10-21): The only server that passed both standard tasks (a live web search and reading a page) in both Claude Code and Cursor, with no API key. Runner-up Fetch passed page reading in both but has no search. DuckDuckGo passed in Cursor but in Claude Code its page reader dropped the page heading; MarkItDown dropped it in both clients. Keyed servers (Brave, Exa, Firecrawl) aren't tested. On Oct 14 Open-WebSearch also passed both tasks in Claude Desktop and in ChatGPT (Codex) (Codex needed a 60 s server start timeout on a cold npx start). MarkItDown failed again in both. Confirmed Oct 21, 2026: a second run passed both tasks again in all 4 clients (open-websearch 2.2.0).

Provisional: based on one dated run per app (no repeat runs yet), so treat it as a candidate, not a final pick. How we test.

What failed or is partial in Cursor

• MarkItDown MCP (Microsoft): failed on 2026-10-21. Recorded reason: no tool result containing ['Example Domain']

Partial: Reddit MCP Buddy. Some steps verified; this is not a pass.

Install in Cursor

The exact snippet we tested is on the Open-WebSearch page (pick the Cursor tab). Not connecting? Cursor MCP troubleshooting.

Not tested in Cursor yet

Apify MCP, Bright Data MCP, Pomerado, SerpApi MCP, Firecrawl MCP, Exa MCP, Tavily MCP, Brave Search MCP, Jina AI MCP, Kagi Search MCP, Xquik, Scrapling, Jevcrawl, Crawl4AI, Xiaohongshu (RedNote) MCP, wigolo, Google Search Console MCP, Open SEO MCP + Skills, designlang (design-extract), FlexQueries, AgentWeb, Awwwards MCP. Listed from their source; no result in Cursor yet.

Questions

What is the best web search & scraping MCP server for Cursor?

In ToolPassed's hands-on tests in Cursor, Open-WebSearch is our provisional candidate for web search & scraping: it passed install, connect, tools and an agent task on 2026-10-21. We ran 5 web search & scraping MCP servers in Cursor; 3 passed every step; not a pass: Reddit MCP Buddy (partial), MarkItDown MCP (Microsoft) (failed).

Does Open-WebSearch work with Cursor?

Open-WebSearch passed in Cursor on 2026-10-21 (Cursor CLI 2026.10.19-6bbf2c4, open-websearch 2.2.0, connected in 2.2s).

Does MarkItDown MCP (Microsoft) work with Cursor?

No, not in our test: MarkItDown MCP (Microsoft) failed in Cursor on 2026-10-21 (Cursor CLI 2026.10.19-6bbf2c4, markitdown-mcp 0.0.1a8, connected in 2.0s). Recorded reason: no tool result containing ['Example Domain']