Extrapify MCP

Extract structured JSON from any public webpage inside Claude Desktop, Cursor, or any MCP-compatible client.

Define a schema. Point it at a URL. Get back validated, typed JSON.

Get an API key → extrapify.com

What this is

A thin, stateless MCP server that bridges MCP clients to the hosted Extrapify API.

Extraction does not happen inside this package. The MCP server forwards requests to the Extrapify API, which handles fetching, Browserless rendering for JS-heavy pages, Claude-powered extraction, schema validation, quota accounting, and observability on the backend.

• MCP protocol server over stdio
• One tool: extract_structured_data
• Production-ready bridge, not a scraping framework
• No extraction logic, no state, no side effects

Install

npm install

Copy .env.example to .env and fill in your credentials:

EXTRAPIFY_API_BASE_URL=https://extrapify.com
EXTRAPIFY_API_KEY=sk_live_your_key_here

Start the server:

npm run mcp:start

Claude Desktop setup

Add this block to your Claude Desktop config (claude_desktop_config.json):

{
  "mcpServers": {
    "extrapify": {
      "command": "node",
      "args": ["/absolute/path/to/extrapify-mcp/mcp/server.mjs"],
      "env": {
        "EXTRAPIFY_API_BASE_URL": "https://extrapify.com",
        "EXTRAPIFY_API_KEY": "sk_live_your_key_here"
      }
    }
  }
}

Restart Claude Desktop. The extract_structured_data tool will appear automatically.

Cursor setup

Cursor supports stdio MCP servers. Point it at node and the local mcp/server.mjs entrypoint with the same two environment variables.

See docs/mcp-install-examples.md for copy-paste configs.

Tool: extract_structured_data

Retrieve structured JSON from any public webpage using a schema you define.

Input:

{
  "url": "https://example.com/article",
  "mode": "auto",
  "schema": {
    "title": "string",
    "author": "string",
    "published_at": "date",
    "tags": "string[]"
  }
}

Output:

{
  "extracted": {
    "title": "How Claude Agents Are Changing Developer Workflows",
    "author": "Jane Smith",
    "published_at": "2026-04-15",
    "tags": ["AI", "agents", "developer tools"]
  },
  "type": "single",
  "count": 1,
  "confidence": 0.96,
  "tokens_used": 1820
}

Supported schema types: string, number, integer, float, boolean, date, datetime, url, and any of these as arrays using [] suffix (e.g. string[]).

Supported mode values:

• auto — let Extrapify decide based on page structure
• single — extract the primary item only
• list — extract all matching items as an array

Schema templates

Starter schemas for common use cases (product pages, job listings, articles, company data) are in docs/schema-templates.md.

Example workflows

Agent patterns and demo workflows are in docs/demo-workflows.md.

Other compatible clients

Any MCP client that supports stdio transport works with this package. Typically you only need:

• command: node
• args: absolute path to mcp/server.mjs
• env: EXTRAPIFY_API_BASE_URL and EXTRAPIFY_API_KEY

Troubleshooting

| Symptom | Fix | |---|---| | Server exits immediately | Verify EXTRAPIFY_API_BASE_URL is a valid absolute URL | | Tool calls return 401 or 403 | Check your API key at extrapify.com/dashboard | | Client cannot discover tools | Confirm it is launching node against mcp/server.mjs over stdio | | Requests time out | Verify the Extrapify API is reachable from your machine | | JS-heavy pages return empty content | Extrapify handles Browserless fallback automatically — no action needed |

Repository layout

mcp/
  server.mjs               ← MCP stdio server entrypoint
  tool-registry.mjs        ← tool definitions
  extrapify-client.mjs     ← minimal Extrapify API client
  tools/
    extract-structured-data.mjs
  configs/
    claude-desktop.local.example.json
    claude-desktop.production.example.json
docs/
  mcp-install-examples.md
  schema-templates.md
  demo-workflows.md
  mcp-marketplace-copy.md

Links

• API and pricing: extrapify.com
• Docs: extrapify.com/docs
• Dashboard: extrapify.com/dashboard