MCP Server

Give AI agents the web.

The Model Context Protocol connects Claude, Cursor, Windsurf, OpenCode, Codex, Antigravity, and any MCP-compatible AI agent directly to webclaw's extraction engine. One JSON config. Full web access. 12 tools your agent can call.

Read full docs

Quick setup

Pick your client. Paste the config. Done.

Claude Desktop

json

{
  "mcpServers": {
    "webclaw": {
      "command": "webclaw-mcp",
      "env": {
        "WEBCLAW_API_KEY": "<YOUR_API_KEY>"
      }
    }
  }
}

Claude Code

bash

claude mcp add webclaw webclaw-mcp

Cursor

json

// ~/.cursor/mcp.json
{
  "mcpServers": {
    "webclaw": {
      "command": "webclaw-mcp",
      "env": {
        "WEBCLAW_API_KEY": "<YOUR_API_KEY>"
      }
    }
  }
}

Windsurf

json

// ~/.windsurf/mcp.json
{
  "mcpServers": {
    "webclaw": {
      "command": "webclaw-mcp",
      "env": {
        "WEBCLAW_API_KEY": "<YOUR_API_KEY>"
      }
    }
  }
}

OpenCode

json

// ~/.config/opencode/opencode.json
{
  "mcp": {
    "webclaw": {
      "type": "local",
      "command": ["~/.webclaw/webclaw-mcp"],
      "enabled": true
    }
  }
}

Codex

toml

# ~/.codex/config.toml
[mcp_servers.webclaw]
command = "~/.webclaw/webclaw-mcp"
args = []
enabled = true

Antigravity

json

// Antigravity MCP config
{
  "mcpServers": {
    "webclaw": {
      "command": "webclaw-mcp",
      "env": {
        "WEBCLAW_API_KEY": "<YOUR_API_KEY>"
      }
    }
  }
}

Installation

Two ways to get the binary.

Pre-built binary

bash

curl -fsSL https://github.com/0xMassi/webclaw/releases/latest/download/webclaw-mcp-aarch64-apple-darwin \
  -o webclaw-mcp && chmod +x webclaw-mcp

From source

bash

cargo install --git https://github.com/0xMassi/webclaw webclaw-mcp

Configuration

All config through environment variables.

Set them in your MCP client config or your shell.

WEBCLAW_API_KEY

Enables cloud fallback for bot-protected sites, handled transparently

OPENAI_API_KEY / ANTHROPIC_API_KEY

Enables extract and summarize tools (LLM features)

OLLAMA_HOST

Use local Ollama for LLM features (default: localhost:11434)

proxies.txt

Place a proxies.txt file in working directory for automatic proxy rotation

12 tools

Everything your agent needs to read, extract, and monitor.

scrape

Extract content from any URL

crawl

Crawl entire sites with BFS traversal

url, depth, max_pages

map

Discover all URLs via sitemap parsing

url, include, exclude

batch

Extract multiple URLs concurrently

urls[], format, concurrency

extract

LLM-powered structured data extraction

url, prompt, schema

summarize

AI-generated page summaries

url, max_length

diff

Track content changes between snapshots

url, previous_snapshot

brand

Extract brand identity from any site

url (returns colors, fonts, logos)

search

Web search via Serper with optional scraping

query, num_results, scrape_results

research

Async deep research with multi-source analysis

query, depth, max_sources

vertical_scrape

Run a site-specific vertical extractor for typed JSON

url, extractor

list_extractors

List every vertical extractor and its output schema

(no params)

Smart Fetch

Local-first architecture: 80% of scrapes cost zero API credits. The cloud API only activates for bot-protected or JS-rendered pages. Your agent gets full web access while keeping costs near zero.

Related guides

How the MCP server fits the wider stack.

Start extracting. Connect in seconds.

Get an API key for cloud features, or run the MCP server fully local with zero configuration.

Read full docs