Cursor + Hydrafetch
Give Cursor today's documentation.
A coding agent writes against the version of a library it learned, not the one you installed. Add Hydrafetch and Cursor reads the current documentation page as clean markdown, headings and code blocks intact, before it writes a line.

Cursor
with Hydrafetch
[ 01 / Set up ]
One block in mcp.json.
Sign in through the browser, or send a key in a header. Both give the agent the same tools.
Sign in through the browser
- 1
Add Hydrafetch to ~/.cursor/mcp.json. Create the file if it does not exist.
~/.cursor/mcp.json{ "mcpServers": { "hydrafetch": { "url": "https://api.hydrafetch.com/mcp" } } } - 2
Open Cursor Settings, then MCP. Hydrafetch asks you to sign in: choose Connect and approve in your browser.
Put the same block in .cursor/mcp.json inside a repo to keep it to that project.
Use an API key instead
- 1
Replace hf_your_key with a key from the API keys page.
~/.cursor/mcp.json{ "mcpServers": { "hydrafetch": { "url": "https://api.hydrafetch.com/mcp", "headers": { "Authorization": "Bearer hf_your_key" } } } }
[ 02 / What to ask ]
Ask for the page. The agent reads it first.
These are the kinds of request that use each tool while you work, with what the call costs.
Read the installation page for this framework and set the project up the way it says.
scrape1 creditFind the migration guide for the version we are upgrading to and apply it.
search1 credit plus 1 per scraped resultList the pages under their API reference so you know what exists.
map1 creditRead the whole guide section of their documentation before you answer.
crawl1 credit per page scrapedPull the request fields from this API page into a typed interface.
extract5 credits per URL
[ 03 / What comes back ]
A documentation page, with its structure.
A real read of one documentation page. 1,519 words came back as markdown with every heading in place and 17 code blocks fenced.
- h1Installation
- h2Quick start
- h2System requirements
- h2Supported browsers
- h2Create with the CLI
- h2Manual installation
- h3Create the `app` directory
- h3Create the `public` folder (optional)
- h2Run the development server
- h2Set up TypeScript
- h3IDE Plugin
- h2Set up your editor
- h2Set up linting
- h2Set up Absolute Imports and Module Path Aliases
- h2Upgrade your Next.js app
the heading outline of the markdown returned, captured from the live API on 2026-11-17
[ 04 / Questions ]
What people ask about Cursor and Hydrafetch.
Add Hydrafetch to Cursor as an MCP server. The agent can then read a documentation page as clean markdown, with its headings and code blocks, before it writes code against that library.
Add a hydrafetch entry with the server address to ~/.cursor/mcp.json, then open Cursor Settings, MCP, choose Connect and approve in your browser. To keep it to one project, put the same block in .cursor/mcp.json inside the repo.
Yes. In a real read of the Next.js installation page on 17 November 2026, the markdown returned had 15 headings and 17 fenced code blocks in 1,519 words.
The same as the API. Reading a page is 1 credit, a search is 1 credit plus 1 per scraped result, and a call that fails costs nothing.
Yes. Add an Authorization header with your key to the same entry in mcp.json. Signing in through the browser and using a key give the agent the same tools.
[ 05 / Keep going ]
Other ways to connect.
[ Start ]
Clean web data is one call away.
500 free credits, no card required. Failures are never billed.
Success rate
Median scrape
ms