OpenBrowserAgent is an AI browser side panel for understanding pages, researching the web, and completing browser tasks with you in control.
It turns your browser into an AI working surface: bring page context into chat, switch between focused Q&A and action-oriented agent mode, connect your own models, and extend the assistant with reusable skills and remote MCP tools.
Custom agents |
Model providers |
Skills |
MCP servers |
Local execution bridges |
Sync |
- Work where the web already is. Ask questions about the current page, compare tabs, summarize content, and keep sources close to the browser session.
- Move from answers to action. Agent mode can inspect pages, navigate tabs, click, type, search, download content, and use richer browser automation tools when needed.
- Bring your own AI stack. Configure model providers, API keys, base URLs, chat models, and image models without being locked into one backend.
- Extend it with skills and MCP. Save repeatable workflows as skills, import skill packages, and connect tested remote Streamable HTTP MCP servers for external tools like search and fetch.
- Stay in control. Tools are visible as they run, MCP servers must be tested before enabling, individual MCP tools can be toggled, and detailed tool JSON is available on demand.
- Side panel chat with streaming responses, queued messages, attachment-aware context, and polished tool activity cards.
- Agent and Ask modes for either browser automation or page-focused questions.
- Tab and page context including current tab metadata, page content, selected elements, and source-aware outputs.
- Remote MCP tools with a built-in web search preset, JSON import, connection testing, per-tool controls, and citation extraction from MCP results.
- Reusable skills with built-in browser guidance, skill import/export, editable skill files, and reset-to-default controls.
- Source citations that keep final answers tied to pages, files, skills, generated outputs, and MCP-provided web results.
- Comprehensive settings for providers, models, appearance, language, sync, agents, skills, MCP servers, and local execution bridges.
- Theme-aware UI with light/dark/system modes, accent colors, compact density, subtle motion, and localized interface text.
- Research a topic from the side panel, fetch pages through an MCP search/fetch provider, and receive a cited summary.
- Ask questions about the active tab or attached pages without leaving the browser.
- Let the agent fill forms, click controls, organize tabs, or inspect page state while showing each tool step.
- Create a custom skill for a recurring workflow, then reuse it across chats.
- Configure multiple model providers and choose the best model for chat or image generation.
npm install
npm run devBrowser-specific development targets are also available:
npm run dev:firefox
npm run dev:safarinpm run build
npm run build:firefox
npm run build:safariBuild outputs are written to .output/chrome-mv3, .output/firefox-mv2, and .output/safari-mv2.