Pagewatch

What the Pagewatch MCP server does

Pagewatch MCP server exposes web-rendering operations to MCP clients. Its three primary tools are read_page, screenshot, and pdf:

• read_page converts a URL into clean Markdown.
• screenshot returns a PNG of the rendered page.
• pdf creates a PDF from either a URL or raw HTML.

The service runs JavaScript while rendering pages. That makes it suitable for sites whose content is not present in the response from a plain HTTP request. It is intended for browser-style access rather than general-purpose scraping logic embedded in the local package.

Two additional tools, register and balance, support key management. On its first call, the service can mint a free trial key and return it with the result. After the trial is used, a human can confirm one email address to unlock 200 additional free credits. The README states that billing is metered but that no payment processor is currently connected, so calls are not charged at present.

How it works

The local package is a thin MCP client for the Pagewatch API. Each tool makes one HTTPS request to a documented /v1 API route; the package does not include a browser, scraping implementation, or credential store. Rendering and conversion are handled by the Pagewatch service.

Pagewatch MCP server can run locally through stdio, or it can use the hosted MCP endpoint at https://pagelens.dev/mcp. The local process can also be configured to serve streamable HTTP by changing its transport setting and providing a port through the runtime environment.

The service respects robots.txt. When a site blocks automation, it refuses the request instead of attempting to bypass that restriction.

Setup and configuration

The documented local setup uses uvx:

{
  "mcpServers": {
    "pagewatch": {
      "command": "uvx",
      "args": ["pagewatch-mcp"]
    }
  }
}

As an alternative, install the package with pip install pagewatch-mcp and configure the resulting pagewatch-mcp command as the MCP server. Docker is also supported through the repository's Dockerfile with docker build -t pagewatch-mcp . followed by an interactive container run.

All configuration variables are optional. PAGEWATCH_API_KEY supplies an existing key; without it, the first call requests a free trial key. PAGEWATCH_BASE_URL changes the API base URL and defaults to https://pagelens.dev. PAGEWATCH_MCP_TRANSPORT defaults to stdio and accepts http for streamable HTTP. PAGEWATCH_SOURCE sets the attribution value recorded during key registration and defaults to mcp-selfhost.

Tools and capabilities

The available operations are:

• read_page: obtain clean Markdown from a URL.
• screenshot: obtain a PNG of a rendered website.
• pdf: convert a URL or raw HTML into a PDF.
• register: manage or create an access key.
• balance: inspect key-related usage information.

Use read_page for agent-readable page content, screenshot when visual output is needed, and pdf when a document artifact is required. The Pagewatch MCP server is especially relevant for JavaScript-rendered pages that a non-rendering fetch cannot populate.

Limitations and notes

The package itself does not contain the browser or page-processing engine; it forwards requests to the Pagewatch API. A reachable Pagewatch deployment is therefore part of the runtime path. Sites that disallow automation are not processed, and the server does not try to evade those controls.

The README describes free trial access and additional free credits after email confirmation, but it does not document credit amounts for the initial trial. It also states that no payment processor is connected currently, so pricing and charging behavior could change independently of the local package. The hosted endpoint is available for users who do not want to run the local process.