Use case · Coding agents

Up-to-date docs for your coding agent

Crawl any docs site into a knowledge base and let Claude Code, Cursor or Codex search it through the SarvCrawl MCP server — current APIs instead of whatever the model remembers.

Start free MCP server docs

21
MCP tools, from kb_crawl to kb_semantic_search
1
command to add the server to Claude Code
1B
free tokens on sign-up, no card

// For developers working with Claude Code, Cursor, Codex, VS Code and Windsurf

IN YOUR EDITOR

Claude · SarvCrawl MCPMCP

Claude CodeCursorVS Code

$ claude mcp add sarvcrawl --env KB_API_KEY=sarv_sk_… -- npx -y sarvcrawl-mcp
> Index the Next.js docs, then tell me how caching works in v16
kb_crawl({
  "kb_id": "kb_nextjs_docs",
  "url": "https://nextjs.org/docs",
  "limit": 400,
  "exclude_paths": ["/docs/pages/*"]
})
kb_check_job_status → completed · 400 pages
kb_semantic_search({ "kb_id": "kb_nextjs_docs", "query": "caching in v16" })
  0.91  Caching and revalidating  /docs/app/guides/caching
  0.87  use cache directive       /docs/app/api-reference/directives/use-cache
  0.82  cacheLife                 /docs/app/api-reference/functions/cacheLife
Answer grounded in 3 current docs pages ✓

Perfect for

Fast-moving frameworks

Index the docs for the version you actually ship, so the agent stops suggesting APIs that were renamed two releases ago.

Internal and private SDKs

Index the docs for your company’s own SDKs, or upload API references as PDFs — docs no model was trained on.

Agent research during a task

Let the agent scrape a changelog or migration guide mid-task and search it, without leaving the editor.

Teams sharing one source

One knowledge base per stack, one API key per developer, the same answers in every editor.

How it works

01 / 06

Add the MCP server

Run one claude mcp add command, or add sarvcrawl-mcp to the MCP config of Cursor, VS Code or Windsurf.

MCP server

02 / 06

Crawl the docs site

Point a crawl at the docs root with a page limit and excluded paths; pages are rendered in a real browser and stored as markdown.

Crawl

03 / 06

Add the PDFs too

Upload specs, whitepapers and API references as PDF or Office files; scans go through OCR.

Parse documents

04 / 06

Search by meaning

The agent calls kb_semantic_search for hybrid keyword-plus-vector results, or kb_search for exact identifiers. Search is on paid plans.

Keyword search Semantic search

05 / 06

Track new releases

Put a monitor on the docs site so changed pages are picked up on a schedule you choose.

Monitor

06 / 06

Check what it indexed

Browse the crawl tree, pages and chunks in the console to see exactly what the agent can find.

Try the playground

How SarvCrawl compares

Feature	SarvCrawl	Pasting docs into chat	Model training data	Editor web fetch
Reflects the current docs
Covers the whole docs site
Search by meaning across pages
Includes your private PDFs
Refreshes on a schedule
Shared across editors and teammates

Stop your agent guessing at APIs

Crawl the docs once, search them from any MCP client. 1 billion free tokens to start.

Start free MCP server docs