29.5K ·GitHub
What is this?
Page Agent is a JavaScript GUI agent that lives inside a web page: one script gives the page its own AI assistant you control with natural language.
Page Agent is a JavaScript GUI agent that lives inside a web page: one script gives the page its own AI assistant you control with natural language.So is it a browser extension, or a script I embed in my own page?A script you embed — plain in-page JavaScript, no browser extension, Python or headless browser needed. An optional Chrome extension exists only for tasks spanning multiple pages.
A script you embed — plain in-page JavaScript, no browser extension, Python or headless browser needed. An optional Chrome extension exists only for tasks spanning multiple pages.How do I add it to a page?Fastest path is a one-line CDN script tag that auto-creates a demo agent; add ?autoInit=false and instantiate new window.PageAgent(...) yourself. For programmatic use, install the NPM package.
Fastest path is a one-line CDN script tag that auto-creates a demo agent; add ?autoInit=false and instantiate new window.PageAgent(...) yourself. For programmatic use, install the NPM package.Which AI models can drive it — do I need a specific provider?No fixed provider: you bring your own LLM — most mainstream models work, including locally deployed ones. The bundled demo LLM is just a free testing API for technical evaluation.
No fixed provider: you bring your own LLM — most mainstream models work, including locally deployed ones. The bundled demo LLM is just a free testing API for technical evaluation.Can I use it for server-side automation, like scraping?No — it is built for client-side web enhancement inside the pages you ship, not for server-side automation.
No — it is built for client-side web enhancement inside the pages you ship, not for server-side automation.What would I actually use this for in a product?Ship an AI copilot inside your SaaS with no backend rewrite, turn 20-click form workflows into one sentence (ERP, CRM, admin tools), add natural-language accessibility, or run agents across tabs.
Ship an AI copilot inside your SaaS with no backend rewrite, turn 20-click form workflows into one sentence (ERP, CRM, admin tools), add natural-language accessibility, or run agents across tabs.Does it look at screenshots to understand the page?No screenshots at all — it reads and manipulates the DOM as text, so it needs no multimodal LLMs and no special browser permissions.
No screenshots at all — it reads and manipulates the DOM as text, so it needs no multimodal LLMs and no special browser permissions.What does the MCP server add?PageAgent ships a beta MCP server so external agent clients can control your browser from outside the page, beyond scripts embedded in the site itself.
PageAgent ships a beta MCP server so external agent clients can control your browser from outside the page, beyond scripts embedded in the site itself.Is it related to browser-use?Yes — the README credits browser-use as the foundation: PageAgent builds upon its work but runs entirely in-page as client-side JavaScript.
Yes — the README credits browser-use as the foundation: PageAgent builds upon its work but runs entirely in-page as client-side JavaScript.How mature is the project?Young but popular: created in September 2025, already over 29,000 GitHub stars, with the MCP server still marked Beta.
Young but popular: created in September 2025, already over 29,000 GitHub stars, with the MCP server still marked Beta.What are similar projects?lcu, playwright-mcp, E2B, midscene, stagehand, mcp-chrome, headroom, OpenIntelligentUI
lcu, playwright-mcp, E2B, midscene, stagehand, mcp-chrome, headroom, OpenIntelligentUIWhich topics is it in?dom, copilot, computer-use, chrome-extension, accessibility, browser-automation, mcp-server, web-development, gui, agent, mcp, npm
dom, copilot, computer-use, chrome-extension, accessibility, browser-automation, mcp-server, web-development, gui, agent, mcp, npm