# OpenCode Browse lets coding agents drive your real, signed-in browser through MCP

Canonical: https://theagenticweekly.com/news/opencode-browse · Evidence: Reported · First seen: 2026-11-09 12:59 UTC · Source published: 2026-11-07 02:35 UTC · Brief published: 2026-11-09 12:59 UTC · Updated: 2026-11-09 12:59 UTC · Author: Agentic Weekly

> OpenCode Browse, announced by OpenCode on November 6, 2026 (US time), is an MIT-licensed MCP server plus Chromium extension that lets any MCP agent drive your real browser with Playwright, using your tabs, logins and extensions, and hand the tab back to you for logins, 2FA and payments. The latest release is v0.0.16 (Nov 8, 2026).

## What changed

- Agents get a browse MCP server that opens and drives tabs with Playwright, takes screenshots, records the screen and captures network traffic as a redacted HAR. The product page says it works with OpenCode, Claude Code, Cursor, Codex, Zed and other MCP clients.
- It runs in your own Chromium browser (Chrome, Edge, Brave, Arc, Helium, Vivaldi, Opera and Chromium). Agent tabs open in the background in a collapsed "Agents" group, and you can lend an agent a tab you already have open with the toolbar button.
- Handoffs: for a login, a 2FA prompt or a payment, the agent hands the tab back and continues only after a real click on Continue. Optional per-session domain allowlists limit which sites an agent's tabs can open.
- Install is one line, npx @opencode/browse install (Node.js 22.19+), or a curl/PowerShell script that installs a single binary. It adds the MCP server to the agents you pick and pairs the extension from the Chrome Web Store with a one-time code.
- The README says the relay listens on 127.0.0.1 only and needs a token, and that agent code runs in a restricted JavaScript realm that "is not a security boundary". OpenCode's dax says it also supports an agent and browsers on different machines.

## Who it is best for

Developers who want a coding agent to check work that sits behind a login: a staging dashboard, an admin panel, a SaaS settings page, or a bug that only shows up in a signed-in session. Headless Playwright setups need you to script logins or copy cookies; Browse reuses the session you already have and hands you the login or 2FA step instead.

It is a poor fit for unattended jobs or for browsers holding banking, email or payment sessions you would not hand to an agent. Pair it with a separate browser profile and a domain allowlist; the README itself warns that page content may try to steer the agent.

## How to try it in five minutes

- Create a separate Chrome or Edge profile for agent work, so your personal logins are not in reach.
- Run npx @opencode/browse install (Node.js 22.19 or newer), tick the agents you use (OpenCode, Claude Code, Codex, Cursor and others) and add the extension when the installer opens the Chrome Web Store.
- Start a session limited to one site, for example browse session new staging --allow-origin staging.example.com, so the agent cannot wander to other hosts.
- Ask your agent: open news.ycombinator.com and tell me the top story. When it reaches a login or 2FA prompt, finish that step yourself and click Continue.
- Revoke access when you are done with browse pair --list and browse pair --revoke <browser>.

## In the source's words

> "it lets your agent control your real browser with your tabs and logins and extensions"
> (OpenCode, launch post on X, https://x.com/opencode/status/2118895128152621153)

> "treat code an agent runs as code running as you."
> (browse README, Security section, https://github.com/anomalyco/browse)

## Early verdict

Worth trying if your agent keeps hitting login walls; run it in a separate browser profile with a domain allowlist, since it drives sessions you are signed into. (Basis: Based on OpenCode's product page, the GitHub README and release notes as read on Nov 9, 2026; we have not run it.)

## What we have not verified

- We did not install or run Browse; features, permissions and security behaviour are as described on opencode.ai, the GitHub README and the release notes.
- We did not test it with Claude Code, Codex or Cursor, or measure how well agents complete real tasks with it.
- The project is at version 0.0.16 and shipped several releases a day over launch weekend, so behaviour and setup steps may change quickly.

## Timeline

- Source published: 2026-11-07 02:35 UTC
- First seen by our watcher: 2026-11-09 12:59 UTC (via X search (daily ops scan): OpenCode launch post)
- Brief published: 2026-11-09 12:59 UTC

## Sources

- [OpenCode Browse product page (opencode.ai)](https://opencode.ai/product/browse) (primary source) (fetched 2026-11-09)
- [anomalyco/browse repository and README (GitHub, MIT)](https://github.com/anomalyco/browse) (primary source) (fetched 2026-11-09)
- [browse v0.0.16 release notes (GitHub)](https://github.com/anomalyco/browse/releases/tag/v0.0.16) (fetched 2026-11-09)
- [OpenCode launch post (X)](https://x.com/opencode/status/2118895128152621153) (fetched 2026-11-09)

## Related

- [opencode: verdict and setup guide](https://theagenticweekly.com/tools/opencode)
- [MCP reference servers: verdict and setup guide](https://theagenticweekly.com/tools/mcp-servers)
- [goose vs opencode](https://theagenticweekly.com/compare/goose-vs-opencode)
- [Alternatives to opencode](https://theagenticweekly.com/alternatives/opencode)
- [All news briefs](https://theagenticweekly.com/news)
- [This week's releases](https://theagenticweekly.com/this-week)
- [New AI models this week](https://theagenticweekly.com/models/this-week)

## How to cite

Agentic Weekly, "OpenCode Browse lets coding agents drive your real, signed-in browser through MCP", published 2026-11-09 12:59 UTC, https://theagenticweekly.com/news/opencode-browse