Agent360 Browser MCP

Your agent needs a 2FA code? It asks you, then carries on in the tab you were already signed into. 40 tools, any client, MIT.

Browser MCP gives Claude Code, Codex, Cursor, VS Code and Windsurf control of the Chrome you already use - the one with your sessions, your cookies, your logins. Not a fresh headless browser that hits a login wall on the first page. ⚠ READ FIRST - THIS EXTENSION IS HALF OF BROWSER MCP Browser MCP is a Chrome extension PLUS a local MCP server. This page can only give you the extension. On its own it will sit on "Not connected" forever, because there is no server for it to talk to. The other half is one command. ═══ SETUP (2 minutes) ═══ 1. Install this extension (you are here). 2. Register the MCP server with your AI client. Claude Code: claude mcp add --scope user browser-mcp -- npx @agent360/browser-mcp@latest OpenAI Codex: codex mcp add browser-mcp -- npx @agent360/browser-mcp@latest Cursor / VS Code / Windsurf - add to that client's MCP config: {"mcpServers": {"browser-mcp": {"command": "npx", "args": ["@agent360/browser-mcp@latest"]}}} Per-client guides: https://browsermcp.dev/docs/install-claude-code/ 3. Restart your AI client, then ask it to use the browser once. A green badge appears on the extension's icon the first time it is actually used - grey before that is normal. Why two steps? Chrome does not allow an extension to install itself from npm, and npm cannot install a Chrome extension. Neither half can install the other, so you install each once. Stuck on "Not connected"? https://browsermcp.dev/docs/troubleshooting/ ═══ WHAT TO SAY ONCE IT WORKS ═══ Nothing happens until you ask. Paste one of these to your agent: "Take a screenshot of my current Chrome tab." Start here - an image back means both halves are talking. "Open my Gmail tab and tell me who sent my last 3 emails." The one that shows the difference: it works because it is YOUR browser, already signed in. A headless tool hits a login wall here. "Go to my analytics dashboard, pull this month's numbers, and put them in a table." Any dashboard you are already logged into - no API key, no export. "Fill in this signup form with my details. Stop and ask me before anything sensitive." You stay in the loop for passwords and payment details. "Log me in here. If it emails a code, read it from my Gmail tab and continue." The move no API can make. The pattern: anything you would do yourself in a browser, on a site you are already signed into. Strongest where there is no API - internal dashboards, admin panels, portals. ═══ WHAT IT CAN DO ═══ 40 tools, including: • Navigate, click, double-click, right-click, hover, scroll, press keys • Read page content, extract lists, take screenshots, read console logs • Fill forms, including React and Vue controlled inputs, dropdowns, comboboxes, date pickers and file uploads • Work inside iframes, switch and close tabs, wait for network activity • Read and write cookies and localStorage, extract API tokens • Detect CAPTCHAs (reCAPTCHA v2/v3, hCaptcha, Cloudflare Turnstile, FunCaptcha): it tries the checkbox, then hands you a screenshot to look at, then asks you to finish it yourself • Ask you directly - an in-page dialog for 2FA codes, passwords and anything else it should not decide alone ═══ TWENTY AGENTS, ONE BROWSER ═══ Up to 20 agent sessions can use one Chrome at the same time. Every chat gets its own colour-coded tab group, can only see its own tabs, and can hold up to 20 of them. Close the chat and its tabs go with it. Other browser automation tools take over the whole profile, so a second agent has to wait. ═══ WHAT IT DOES WITH YOUR DATA ═══ Nothing is sent to us. The extension talks to one thing: a server on 127.0.0.1 that you started. There is no account, no telemetry, no server of ours, and no payment. What your agent reads goes to your AI client and its model provider, the same as anything else you show it. The extension keeps a local log of every action the agent took, and marks the sensitive ones, so you can look at what happened. ═══ WHAT IT CANNOT DO ═══ In a tab you are not looking at, the agent can navigate, read, screenshot, run scripts, click, and fill a field it finds by CSS selector. It cannot press keys, hover, double-click, right-click, click by coordinate, type into a combobox, or fill a field found by its visible text - Chrome does not deliver mouse and keyboard events to a tab that is not the visible one in its window. When one of those is needed, the tool says so and brings the tab in front of you rather than failing silently. This is a Chrome limitation, not a bug we are about to fix. Checkout https://computermcp.dev for full access like a human e2e MIT licensed, source at https://github.com/Agent360dk/browser-mcp