Install

Pick the surface that matches the job:

• Pasting the page you are reading into a chat → the browser extension (Chrome, Firefox).
• An agent reading public pages (Claude Code, Cursor, Codex, any MCP client) → the MCP server.
• ChatGPT reading a public page → the ChatGPT plugin.
• HTML that you or your agent already loaded (your own app, a page behind a sign-in, an intranet or tailnet host) → the MCP tool’s html argument, or the library.
• An agent that drives a browser (Chrome DevTools MCP, Playwright MCP) and wants the extension’s Markdown of the page in it → the agent script.

What each one cannot do: Which Page2AI do I need?

Chrome extension (Manifest V3)

The recommended way for humans clipping one page at a time.

Chrome Web Store: live since 2026-07-30. Install it from the store listing. To see the published version yourself, no account needed, ask the endpoint a real browser uses for extension updates:

curl -s 'https://clients2.google.com/service/update2/crx?prodversion=140.0&acceptformat=crx2,crx3&x=id%3Ddlpaaijcnbbmlfeohlphjpnbbcnomnno%26uc'

status="ok" plus a version means it is live. error-unknownApplication means it is not.

Install from a release. Every tagged version ships a zip built by CI on the releases page. Unzip it and load it unpacked via chrome://extensions → Developer mode → Load unpacked.

Build from source (for reviewers / local dev).

git clone https://github.com/igorsaevets/page2ai-extension
cd page2ai-extension
npm install
npx wxt zip -b chrome

Output lands in .output/page2ai-extension-<version>-chrome.zip. Load unpacked from .output/chrome-mv3/ via chrome://extensions → Developer mode → Load unpacked.

Trigger extraction. Press Alt+Shift+M on any page, or click the toolbar icon, to open the popup, then press Enter or select Extract Markdown. When it is done, the Markdown is on your clipboard.

Firefox extension (Manifest V3)

Firefox Add-ons (AMO): listed since 2026-07-31. Install it from the AMO listing. To see the published version yourself, no account needed: curl -s "https://addons.mozilla.org/api/v5/addons/addon/page2ai@igorsaevets.com/" and read current_version.version.

Build from source:

npx wxt zip -b firefox

Load in about:debugging → This Firefox → Load Temporary Add-on → pick manifest.json from .output/firefox-mv3/.

MCP server (agents)

For Claude Desktop, Cursor, Windsurf, Zed, and any other MCP client.

One-liner install. Add this to your MCP client config:

{
  "mcpServers": {
    "page2ai": {
      "command": "npx",
      "args": ["-y", "page2ai-mcp"]
    }
  }
}

The unscoped page2ai-mcp package is a thin wrapper that reexports @page2ai/mcp; both are on npm.

Verify. From a shell:

npx -y page2ai-mcp --version

It prints the version and exits (--version needs 0.6.2 or later). It needs Node.js 20 or later.

Without --version the server starts on stdio, prints [page2ai-mcp] server <version> ready on stdio to stderr and waits for an MCP client: that is normal, Ctrl+C to exit. The lines about Agent Bridge that follow it are about connecting that server to extension 1.6.0 — ignore them unless you are setting up the Agent Bridge below. To see the server answer a request, run the initialize example.

Agent Bridge (extension 1.6.0). To let the server read pages through your browser (behind a sign-in, or rendered by JavaScript): open the extension popup, turn on Agent Bridge, allow access to websites, and give the server the key the popup shows — full steps and the version table are on the MCP server page. If your store install has no such switch, update the extension first, then set up the bridge; otherwise pass the page’s HTML as the tool’s html argument instead.

Full details: MCP server page.

ChatGPT (plugin)

Page2AI is in ChatGPT’s plugin directory: Page2AI · Public web page to Markdown. Page2AI itself needs no account or key.

1. In ChatGPT on the web, desktop or mobile, open Plugins in the sidebar and find Page2AI, or open the link above.
2. Select Install plugin.
3. In a chat, give ChatGPT a page address and ask for the page as Markdown. ChatGPT uses an installed plugin when it fits the request; to pick it yourself, type @ and choose Page2AI, or select + and enter its name.

Whether you can install a plugin depends on your plan and, in a workspace, on what its administrators allow (OpenAI Help: Plugins in ChatGPT).

The plugin reads public pages only. It calls the hosted endpoint, which fetches the URL from its own server, without your cookies, and runs no page scripts: a page behind a sign-in comes back as its login page, and a page that JavaScript builds in the browser comes back nearly empty. For those, copy the page with the extension and paste it into the chat. The plugin takes a URL only and waits at most 20 seconds for a page.

Agent script (browser automation)

For an agent that already drives a browser through Chrome DevTools MCP or Playwright MCP. The script is the extension’s own extractor as a plain page script: the agent gets the Markdown the popup would give for the page that browser shows, a page behind a sign-in too when that browser is signed in, without installing the extension. It reads only the page and makes no network request.

Every release since 1.5.0 ships it on the releases page:

• page2ai-agent.fn.js for Chrome DevTools MCP (evaluate_script with sourcePath);
• page2ai-agent.js for Playwright MCP (--init-script, then browser_evaluate);
• page2ai-agent.SHA256SUMS: check both files with sha256sum -c page2ai-agent.SHA256SUMS.

The result is one JSON object with the Markdown in markdown. Have the browser server write it to a file (filePath in Chrome DevTools MCP, filename in Playwright MCP), then jq -r .markdown result.json > page.md. Steps for both servers: agent/README.md.

npm library (Node or browser code)

For programmatic HTML → Markdown conversion in your own code.

npm install @page2ai/core

import { htmlToMarkdown, fetchAndConvert } from '@page2ai/core';

const html = '<article><h1>Hello</h1><p>World</p></article>';
const { markdown, title } = htmlToMarkdown(html, {
  baseUrl: 'https://example.com/hello',
  includeFrontmatter: true,
});

const result = await fetchAndConvert('https://docs.anthropic.com/en/api/messages');
console.log(result.markdown);

Both htmlToMarkdown (sync) and fetchAndConvert (async, with SSRF guard + timeout) are exported.

Node 18.17 or later. Zero external API calls; uses linkedom for parsing.

---

Written by Igor Saevets — LinkedIn · GitHub.