OpenCode Browse Gives AI Agents Your Signed-In Browser

OpenCode has launched Browse, an MCP server and Chromium extension that hands AI coding agents the keys to your real browser session, complete with signed-in tabs, saved logins, and active extensions. Officially released on November 17, 2026, Browse eliminates the headless workarounds and API-key juggling that have defined browser automation, letting agents work inside the same Chrome, Edge, Brave, Arc, or Vivaldi window you use every day. A Hacker News post announcing the launch hit 148 points in six hours, signaling strong developer interest in the approach.
How OpenCode Browse Works
Browse installs as both a Chromium extension and an MCP server. Once configured, coding agents open new tabs in collapsed background groups, keeping them out of sight while they read pages, click buttons, type text, capture screenshots, and record tab activity. When an agent encounters a login prompt, a two-factor authentication challenge, or a payment screen, it hands the tab back to the user. You complete the action, click Continue, and the agent resumes where it left off. The system preserves your existing cookies, local storage, and extension state, so agents operate with the same privileges and context you have.
Supported Browsers and Clients
Browse works with any Chromium-based browser, including Chrome, Microsoft Edge, Brave, Arc, and Vivaldi. On the agent side, it integrates with every MCP client: Claude Code, Cursor, Codex, and Zed all gain browser-control capabilities through the same server. The installation process begins with the Chrome Web Store extension, which walks users through CLI setup and MCP configuration. Developers who prefer manual installation can bypass the extension and install the command-line tool directly, though OpenCode advises against it for most workflows.
Availability and Installation
Officially launched on November 17, 2026, Browse is available now through the Chrome Web Store and the OpenCode CLI. The extension guides users through a step-by-step setup that links the browser extension to the MCP server, ensuring agents and browsers communicate over a local socket. No API keys are required, and no cloud service sits between the agent and the browser. The entire stack runs locally, with agents issuing commands through the MCP protocol and the extension executing them in real tabs.
What This Means
OpenCode Browse removes a longstanding barrier in AI-assisted development: agents that need to interact with web UIs have traditionally relied on headless browsers or Puppeteer scripts that cannot access your logged-in sessions. Browse flips that model, treating your everyday browser as the execution environment. The result is a workflow where agents can file GitHub issues, test deployed applications, scrape authenticated dashboards, and handle multi-step web tasks without forcing developers to manage separate credential stores or mock authentication flows. As MCP adoption grows across coding tools, Browse sets a new baseline for how agents should interact with the web.
on Emergent today
.avif)




