Fetcher MCP

What jae-jae/fetcher-mcp MCP server does

The jae-jae/fetcher-mcp MCP server retrieves web pages through Playwright and is intended for sites whose content depends on JavaScript execution. It can return either the page's main readable content or the broader page content. By default, extracted content is returned as Markdown; callers can request HTML instead.

The server exposes three tools:

• fetch_url loads one URL and returns its content.
• fetch_urls loads several URLs concurrently using multiple browser tabs and combines the results with separators.
• browser_install installs the Chromium binary required by Playwright.

Content extraction uses a Readability-based approach to remove elements such as navigation and advertisements. Resource blocking can disable images, stylesheets, fonts, and media to reduce unnecessary loading.

How it works

Each fetch runs in a browser context rather than using a basic HTTP request. The request can specify when navigation is considered complete: load, domcontentloaded, networkidle, or commit. The default navigation timeout is 30 seconds, while an optional second navigation wait can handle redirects or verification flows.

fetch_url accepts a required url and optional controls for timeout, extraction, output type, maximum character count, media blocking, and debug mode. fetch_urls accepts a required array of URLs and supports the same fetch settings. The default behavior extracts the main content, returns Markdown, imposes no content-length limit, and disables media resources.

For sites that require an interactive login, jae-jae/fetcher-mcp MCP server can run with a visible browser window. A user can log in manually, after which the authenticated browser session is used for fetching. Debug mode can be enabled when starting the process or for an individual request.

Setup and configuration

Run the server locally with:

npx -y fetcher-mcp

Playwright Chromium must be installed separately the first time:

npx playwright install chromium

The documented Claude Desktop configuration starts the server over standard input/output with the npx command and the -y fetcher-mcp arguments. The project also provides a Docker image at ghcr.io/jae-jae/fetcher-mcp:latest and documents port 3000 for container deployment.

For network-based clients, start jae-jae/fetcher-mcp MCP server with --transport=http. This provides a Streamable HTTP endpoint at /mcp and an SSE endpoint at /sse; host and port can be set with --host and --port. The --debug flag displays the browser window.

Limitations and notes

A working Chromium installation is required for browser operations. Some pages may need longer timeouts, an additional navigation wait, or visible-browser interaction for verification and login steps. Intelligent extraction may omit content that is not identified as the page's main content; set extractContent to false when the complete page is needed. Set returnHtml to true when downstream processing requires HTML rather than Markdown.

The README documents local execution, HTTP/SSE transport, Docker deployment, and Claude Desktop configuration. It does not document API-key authentication or server-side credentials.