Webclaw

What 0xMassi/webclaw MCP server does

The 0xMassi/webclaw MCP server exposes web-content operations through the Model Context Protocol. Its core extraction path can turn a URL into markdown, plain text, JSON, cleaned HTML, or a compact format intended for language-model context. The JSON output can include structured metadata, links, images, and extracted fields.

The tool set covers both one-page and multi-page work. scrape handles a single URL, crawl follows same-origin links, and map discovers URLs without extracting every page. batch processes multiple URLs in parallel. diff compares a current page with a saved content snapshot, while brand identifies colors, fonts, logos, and metadata. extract converts page content into structured data, and summarize produces a page summary.

How it works

The project includes an MCP server, CLI, extraction engine, REST API, and SDKs. MCP-compatible clients invoke the server’s tools, while the CLI and API provide other ways to use the same project. Local processing is available for the core extraction tools and does not require an account. The README also describes TLS fingerprinting for working around some anti-bot protections without a browser.

extract and summarize work locally with a local or configured large language model. search and research use the hosted API, rather than the local-only extraction path. The hosted service is intended for cases such as protected-site access, JavaScript rendering, asynchronous crawl or research jobs, web search, watches, and production usage tracking. These hosted capabilities require a WEBCLAW_API_KEY.

Setup and configuration

The 0xMassi/webclaw MCP server can be added to a client with the package launcher below:

{
  "mcpServers": {
    "webclaw": {
      "command": "npx",
      "args": ["-y", "@webclaw/mcp"]
    }
  }
}

The repository says that npx create-webclaw can detect supported AI tools and write their MCP configuration. Named clients include Claude Code, Claude Desktop, Cursor, Windsurf, OpenCode, and Codex CLI. The project also provides Homebrew installation, downloadable binaries, Docker usage, and Cargo commands for its CLI and MCP packages.

Set WEBCLAW_API_KEY when routing requests through the hosted webclaw API or using hosted capabilities. The local core path can run without this variable or an account. The repository also documents TypeScript, Python, and Go SDKs for application-level use, but those SDKs are separate from configuring the MCP client.

Tools and capabilities

• scrape: return one page in markdown, text, JSON, LLM, or HTML format.
• crawl: follow same-origin links and extract discovered pages.
• map: enumerate URLs without extracting each page.
• batch: scrape several URLs in parallel.
• extract: produce structured data from page content.
• summarize: summarize a page.
• diff: compare page snapshots.
• brand: collect brand colors, fonts, logos, and metadata.
• search: search the web and scrape results through the hosted API.
• research: run a multi-source research workflow through the hosted API.

Limitations and notes

Local and hosted capabilities are not identical. Search and research are marked as hosted-only, while most extraction tools are available locally. JavaScript rendering, protected-site access, asynchronous jobs, web search, watches, and production usage tracking are listed as hosted-service use cases rather than local features. The 0xMassi/webclaw MCP server is therefore a better fit for local page extraction when account-free operation matters, and for hosted operation when those additional web-access features are needed.