Skip to content

New: deep research, balance and auto-reload tools in the keyless MCP server. Install in one line

/mcp

Web search for your agent, in one line

A keyless MCP server for Claude Code, Cursor, Windsurf and any MCP client: 6 tools for search, reading pages, deep research and paying its own way.

terminal

$ claude mcp add contextfetch -- npx -y @contextfetch/mcp

contextfetch_search
{ "query": "python asyncio timeout", "num_results": 5 }
  1. 1The asyncio Timeout Trap: Why wait_for() Doesn't Stop Tasks Instantly and How to Fix Itrunebook.dev
  2. 2GitHub - aio-libs-abandoned/async-timeout: asyncio-compatible timeout class · GitHubgithub.com
  3. 3The async with timeout: Best Practices for Robust Python Asynchronous I/Orunebook.dev
  4. 4Python asyncio.wait_for(): Cancel a Task with a Timeoutpythontutorial.net

Works with

Any client that runs stdio MCP servers. No signup, no key until you need more.

  • Claude CodeMCP, one line
  • CursorMCP server
  • WindsurfMCP server
  • exa_pychange base_url
  • tavily-pythonchange api_base_url
  • Any MCP clientstdio, keyless

Tools purpose-built for your agents

Every tool returns Markdown your model can read directly, with a payment link instead of a crash when credits run out.

  • 1 credit

    contextfetch_search

    Search the web; each result comes back as Markdown with only the relevant passages.

    {query, num_results = 5, include_domains?, exclude_domains?}

  • 1 credit

    contextfetch_extract_url

    Read any public page as Markdown, or only the passages that answer a question.

    {url, query?}

  • 1 credit

    contextfetch_find_similar

    Pages like a given URL, outside its domain: alternatives, related docs.

    {url, num_results = 5}

  • 1 credit per sub-query

    contextfetch_deep_research

    Up to 3 searches in parallel, deduped into one Markdown evidence pack.

    {query, sub_queries?, num_results_per_query = 5}

  • Free

    contextfetch_balance

    Credits left, the 7-day burn rate and auto-reload settings.

    {}

  • Free

    contextfetch_auto_reload

    Turn auto-reload on, off or change it, with a monthly cap of at most $500.

    {enabled, threshold?, pack?, monthly_max_usd?}

New tools

Research and billing, without leaving the chat

contextfetch_deep_research
{
  "query": "How do asyncio timeouts cancel tasks?",
  "sub_queries": [
    "asyncio.timeout vs wait_for",
    "asyncio CancelledError cleanup",
    "python 3.11 timeout context manager"
  ]
}

→ one Markdown evidence pack · 3 credits

Deep research in one call

Your agent passes up to three angles on a question and gets one deduped evidence pack with titles, URLs, scores and pruned context.

contextfetch_auto_reload
{
  "enabled": true,
  "threshold": 1000,
  "pack": "builder",
  "monthly_max_usd": 500
}

✓ monthly cap ≤ $500 through MCP

Credits that top themselves up

Your agent can check its balance and turn on auto-reload: below 1,000 credits, buy Builder again, capped per month.

Install

One line in Claude Code

Or add the same npx command to any client's MCP config. Set a key only when you want more than the free tier.

  • Keyless by default: 25 calls a day per IP
  • CONTEXTFETCH_API_KEY for a paid key
  • Published on npm: @contextfetch/mcp

MCP reference

claude mcp add contextfetch -- npx -y @contextfetch/mcp # with a paid keyclaude mcp add contextfetch -e CONTEXTFETCH_API_KEY=YOUR_API_KEY -- npx -y @contextfetch/mcp

Pricing

MCP server for your agents

$1.00/ 1k tool calls

1 credit per search, page or find-similar call. Balance and auto-reload are free. One-time payment. Credits never expire. Full Markdown and pruning included.

Questions

Do I need an API key?

No. Without CONTEXTFETCH_API_KEY the server runs keyless: 25 requests per IP per UTC day. Set the variable to a paid key for more. When credits or the free tier run out, tools return the payment link as an error result instead of crashing, so your agent can show it to you.

Which clients does it work with?

Any client that runs stdio MCP servers: Claude Code, Claude Desktop, Cursor, Windsurf and others. It is an npm package started with npx, so Node.js is the only requirement.

What does each tool cost?

Search, extract_url and find_similar cost 1 credit per call. deep_research costs 1 credit per sub-query, at most 3. balance and auto_reload are free.

Can my agent spend money on its own?

Only through auto-reload on a key whose card was saved at checkout. Through the API or MCP a key can turn auto-reload off, lower its values or change the pack, but the monthly cap can never go above $500; a higher cap needs the browser. A declined payment turns auto-reload off and the balance tool reports the error.

Where does it send requests?

To https://api.contextfetch.dev, or the URL in CONTEXTFETCH_API_BASE. The key travels only in the Authorization header, and we store only its SHA-256 hash.

Add the web to your agent

One line, no key, no signup.