Servo Fetch

What konippi/servo-fetch MCP server does

konippi/servo-fetch MCP server exposes web browsing and extraction functions through Model Context Protocol. It can retrieve a page after rendering JavaScript and CSS, return cleaned Markdown or structured JSON, capture a PNG screenshot, execute supplied JavaScript, and process multiple URLs. The underlying binary embeds the Servo browser engine rather than relying on Chromium or a hosted browser service.

The project is useful for agents that need more than raw HTTP responses. Rendered-position and visibility-aware extraction can remove common layout noise such as navigation, sidebars, footers, cookie notices, modals, and CSS-hidden elements. Schema-based extraction uses CSS selectors to produce structured records from repeated page elements.

How it works

The CLI starts the MCP interface with servo-fetch mcp. A local stdio configuration invokes the binary directly, while servo-fetch mcp --port 8080 enables the streamable HTTP mode. The MCP server provides six tools: fetch, batch_fetch, crawl, map, screenshot, and execute_js.

Fetching uses Servo to run page JavaScript and calculate CSS layout. Screenshots use a software renderer, so a GPU is not required. Batch fetching handles several URLs concurrently. Crawling follows links with breadth-first traversal, same-site scope, robots.txt support, and rate limiting. URL mapping uses sitemaps without rendering pages, which provides a lighter discovery path.

Browser sessions use one-use worker processes, keeping cookies and browser storage isolated between sessions. Requests can also include cookies and custom headers through the CLI and underlying APIs.

Setup and configuration

Install the CLI with the project’s shell installer, or use a prebuilt release, cargo binstall servo-fetch-cli, or cargo install servo-fetch-cli. The README also documents library installations for Rust, Python, and Node.js, but those are separate interfaces rather than requirements for the MCP process.

A stdio client configuration uses the following command and argument:

{
  "mcpServers": {
    "servo-fetch": {
      "command": "servo-fetch",
      "args": ["mcp"]
    }
  }
}

Linux headless environments may need libegl1, libfontconfig1, and libfreetype6, along with xvfb-run. macOS requires no extra setup according to the README. On Windows, the release ZIP is recommended because it includes the required libEGL.dll and libGLESv2.dll sidecar files.

Tools and capabilities

• Fetch pages as cleaned Markdown or structured JSON.
• Fetch multiple URLs concurrently with batch_fetch.
• Crawl a site within a same-site boundary and configured limit.
• Discover URLs from a sitemap with map.
• Capture page or full-page PNG screenshots.
• Run JavaScript in the page context with execute_js.
• Extract fields from CSS-selector schemas.
• Work through the CLI, Rust library, Python SDK, Node.js SDK, or HTTP API in addition to MCP.

Limitations and notes

konippi/servo-fetch MCP server requires the local servo-fetch executable for the documented stdio setup. The README states that it does not require an API key, but it does not describe authentication, access controls, or deployment security for the HTTP mode. Browser behavior is provided by Servo, so compatibility with every website or browser feature is not established by the supplied material.

The project lists Python 3.11 or later for its Python SDK. Linux deployments may need the additional graphics and font libraries noted above. The documented license is MIT OR Apache-2.0.