Agent Browser
Agent Browser is a fast, native Rust CLI that enables AI agents to drive Chrome via the DevTools Protocol. It offers stable element references, structured read/snapshot tooling, network interception, accessibility audits, React introspection, and a typed MCP server for safe, reviewable automations.
Overview
Typical flows open a browser, take a snapshot to obtain stable element references, and then perform actions by ref (click, fill, press). Batch execution reduces per-command startup overhead for long sequences. Deterministic errors surface blockers like overlays, improving agent reliability and recovery.
Capabilities and Architecture
Agent Browser suits agent framework developers, QA engineers, SREs verifying SSR/CSR states, growth and onboarding teams testing funnels, technical writers building docs bots, and support engineers reproducing authenticated issues. It’s ideal when you need stable, observable actions, strong error signals, and programmable guardrails for production-grade agent automations.
- Generate stable element references from snapshots to reduce brittle CSS selectors.
- Fetch readable markdown or text and outline pages for documentation-aware navigation.
- Intercept, mock, and record network traffic with routing controls and HAR capture.
- Run offline accessibility audits and React profiling with structured, automatable output.
- Expose typed MCP tools with allowlisted domains for safe, approvable automations.
Why It Matters
Field-Proven Reliability Signals
Install globally with your preferred package manager or use Homebrew or Cargo, then run install to fetch Chrome for Testing. Open a session, snapshot, and act by ref for stable interactions. Enable the React DevTools hook at launch to use React tree, renders, and Web Vitals. Start the MCP server to expose typed tools to clients. Persist auth via profiles, sessions, or saved state files; import cookies when needed. Use doctor for environment checks and upgrade for seamless updates. Batch mode consolidates multi-step flows into one invocation, minimizing overhead.
Stable refs, explicit failure modes, and typed tools turn browser control into a dependable automation primitive.
Getting Started and Setup
Agent Browser stands out by combining a native CDP core with snapshot-based refs, docs-aware reading, and typed MCP tools. Deterministic errors surface blockers early, while audits, diffs, and profiles create trustable evidence. It avoids heavyweight dependencies, supports multiple install paths, and keeps advanced features available offline under strict security policies.
Open the tool and review its core product experience.
Create your account or access your existing workspace.
Use your own task to judge speed, quality, and fit.
Check similar AI tools before making a final decision.
Agent Browser
Vercel Labs explores advanced developer and AI tooling that complements the modern web stack. The team builds fast, pragmatic, open-source experiments and infrastructure that emphasize reliability, observability, and great ergonomics for agents, frameworks, and human operators.
Agent Browser is a native Rust command-line tool for reliable, scriptable browser automation designed for AI agents. It launches or connects to Chrome for Testing and exposes a rich, typed surface over the Chrome DevTools Protocol. Agents work from snapshots that return stable element references, reducing selector fragility and improving reproducibility. Semantic finders, traditional selectors, screenshots, PDFs, diffs, and batch execution enable complex, multi-step workflows with minimal overhead. A powerful read command fetches agent-friendly markdown, outlines, and nearest llms metadata for documentation-aware navigation. Built-in network interception, HAR capture, cookies/storage management, and state save/load streamline authenticated and deterministic runs. Accessibility audits use an embedded axe-core engine; React introspection and Web Vitals are available when launched with the React DevTools hook. An MCP server exposes typed tools with allowlisted domains and approval prompts, making automations safer and easier to supervise.
github.com/vercel-labs/agent-browser
Comments (0)
No Comments Found