Rank #05 · Weekly

Rank #05 · Weekly · Agent-Skill

agent-browser: Vercel Labs' Rust Browser Automation CLI for AI Agents

· Weekly Top10 · License: Apache-2.0

Rank #05 Weekly — vercel-labs/agent-browser

Browser computer-use for any agent: snapshot returns compact refs like @e1, the agent clicks or fills, then snapshots again — one short shell call per step.

Canonical digest: /daily/agent-browser (also on Weekly board).

  • Board: Rank #05 · Weekly ·
  • Repo: vercel-labs/agent-browser
  • Stars / lang / license: 51,631 stars · Rust · Apache-2.0
  • License label: License: Apache-2.0
  • Data verified: 2026-11-16 09:10 Asia/Shanghai (official Weekly Top10 gate; language/license confirmed via GitHub API get_repository + prior README notes)
  • Trending note: +4,217 stars this week (weekly Trending; selection captured 09:10 Asia/Shanghai)
  • Tags: Agent-Skill · browser-automation · computer-use · Rust · CLI · MCP · accessibility · Vercel

What it does

agent-browser is a native Rust CLI (npm-distributed) that lets an AI agent drive a browser one command at a time. The loop:

  1. Open a page (agent-browser open example.com).
  2. Snapshot (agent-browser snapshot -i) — page as compact accessibility-tree text with a ref per interactive element.
  3. Act on a ref (click @e1, fill, type, …), then snapshot again to verify.

Design points from the README (as verified):

Who it's for

Less ideal if you need native desktop apps (see Vercel Labs' agent-desktop) or only need a classic Playwright test suite with no agent loop — Playwright remains a fine fit for that.

Stars / language / license

51,631 stars · Rust · Apache-2.0

Data verified: 2026-11-16 09:10 Asia/Shanghai (official Weekly Top10 gate; language/license confirmed via GitHub API get_repository + prior README notes)

Trending note: +4,217 stars this week (weekly Trending; selection captured 09:10 Asia/Shanghai)

Category tags

Agent-Skill · browser-automation · computer-use · Rust · CLI · MCP · accessibility · Vercel

How it differs from similar projects

Compared with Playwright / Puppeteer / Selenium. Those are general automation SDKs aimed at tests and scripts. agent-browser is shaped as an agent tool: compact accessibility snapshots, ref handles, shell-sized steps, MCP/skills packaging, and session/security affordances for untrusted agent loops.

Compared with vision computer-use (screenshot → click coordinates). Structure-first refs vs pixel guessing. Accessibility metadata quality still limits what the agent can see.

Compared with chrome-devtools-mcp (on today's Daily board, not this weekly list). That official Chrome DevTools MCP server emphasizes DevTools debugging, performance traces, network/console inspection, and Puppeteer-backed automation. agent-browser's documented loop is snapshot → ref → act. They can overlap on driving Chrome; TidyHow has not compared coverage or reliability, and this is not a parity claim.

Compared with Vercel Labs' agent-desktop. Same family and loop; agent-browser stops at the browser, agent-desktop drives native OS windows. agent-desktop is not on this weekly board.

No independent accuracy or speed benchmarks are claimed here.

Repo link

https://github.com/vercel-labs/agent-browser