AI Agent Index
ByHeather MacAvelia·Independently reviewed·Published May 14, 2026·Updated Sep 28, 2026
Independently verified against live vendor data on Sep 7, 2026.

Open-source MIT-licensed Python library for AI browser automation, plus a pay-as-you-go cloud with managed browsers from $0.02 per browser-hour. MCP server and client.

How we scored it

Autonomy

5/5

Integrations

4/5

Pricing clarity

4/5

Evidence

5/5

Setup

5/5

The facts

Browser Use is an open-source Python library that lets AI agents control a real browser using LLMs instead of brittle CSS selectors or XPath scripts. Rather than pre-programming every interaction with selector code, a developer describes the task in natural language and the agent works out how to execute it against whatever website it encounters, including sites it has never seen before. The repository holds 117k GitHub stars and carries an MIT license, so the library itself is free to run on your own compute with your own LLM API keys. Because the agent reasons from the current visual and DOM state rather than hardcoded selectors, it adapts when a website layout changes, which makes it usable against legacy web applications and any site without a public API. Browser Use, Inc. sells two commercial products on top of the library. Browser Use Agents take a task and return completed work through the V4 API, with GPT-5.6 Luna as the default model and Claude, Grok, MiniMax and Gemini models selectable per run. Browser Infrastructure gives your own agents and automations managed cloud browsers through the SDK, REST or the Chrome DevTools Protocol, with stealth fingerprinting, free CAPTCHA solving, persistent profiles and residential proxies in 195 plus countries. Playwright, Puppeteer and Selenium scripts connect to those browsers over CDP. The library integrates with any LLM via BYOK, including OpenAI, Anthropic, Google Gemini, Groq, OpenRouter, Azure, Bedrock, DeepSeek and local models through Ollama, and the docs carry examples for LangChain and for n8n, Make.com and Zapier orchestration. Browser Use ships a local MCP server over stdio and a hosted Cloud MCP server at api.browser-use.com, documented for Claude Code, Claude Desktop, Cursor and Windsurf. It also acts as an MCP client: the library includes a client module, and Browser Use Terminal takes an MCP config so the agent can call tools on external MCP servers. Named gaps: the cloud has no no-code builder for non-developers, content behind OAuth walls needs a saved browser profile or a human hand-off for the login step, and internal SaaS tools need custom tool definitions. Pricing is public, in US dollars, and has no subscription. Every account starts free with a one-time $15 credit for eligible Google, GitHub or Microsoft signups, ten concurrent sessions and access to the cost-efficient models only. Paid usage is bought as credits, from $5 to $100,000 per top-up, and the credits never expire. Cloud browser time is $0.02 per browser-hour metered by the minute with a one-minute minimum, managed residential proxy traffic is $5 per GB, and direct or bring-your-own-proxy traffic is $0.20 per GB. Browser Use Agents bill the model cost plus 20 percent, so $1 of model usage costs $1.20 on top of browser time and traffic. Concurrency rises with lifetime spend, from ten sessions at $0 to 1,000 at $25,000, and Enterprise adds a negotiated annual credit pool, SLAs, data-retention terms and dedicated support. The Dev, Business and Scaleup subscriptions that were sold until at least July 2026 are no longer offered to new accounts, and existing subscribers keep their plan. What the free account does not buy is specific. The $15 credit does not renew, all models unlock only after a first top-up, and auto-recharge needs a card on file. The privacy policy, last updated July 15, 2026, states that Browser Use uses Inputs to train its AI models, and the only opt-out it describes covers marketing email. Teams that cannot accept that run the open-source library on their own infrastructure, where no cloud data leaves the machine, or negotiate data-retention terms on an Enterprise plan. Browser Use is SOC 2 Type II compliant, audited by Accorp Partners over a July 17 to October 17, 2025 observation window, with the report available on request to support. No HIPAA, GDPR or ISO 27001 certification is published on any Browser Use surface. Current state Q3 2026: Browser Use Agents V4 is the current API for new integrations, with API V3 and V2 still supported. The vendor publishes its own benchmarks, including 82 percent on a 106-task internal set at $0.17 per solved task, 98 percent across 300 Online-Mind2Web tasks, and 81 percent on a 71-site stealth benchmark, all measured by Browser Use itself. Browser Use Box, a 24/7 remote VM with Claude Code and Browser Harness pre-installed, launched in April 2026, and Browser Use Desktop is an open-source desktop app in its own repository. Hermes Agent lists Browser Use as its default browser tool. The G2 profile exists but holds no reviews, so the independent evidence rests on the GitHub repository at 112,000 plus stars.

Pricing

freemium · $0.02 per browser-hour

View pricing ↗

Segment

b2b

Setup

easy

Verified

Sep 7, 2026

Transparency

Mostly Public

Contract

—

Data training

Trained

Human in loop

Optional

Capabilities

web-searchdata-analysisautonomousno-codeworkflow-builder

Pros & Limitations

Editorial assessment

Pros

  • ✓Self-healing browser automation using LLMs and computer vision instead of brittle CSS selectors or XPath, adapting to website layout changes without code maintenance and making it reliable for automating legacy web apps, government portals, and sites without public APIs.
  • ✓MCP support in both directions. A local stdio server and a hosted Cloud MCP server at api.browser-use.com are documented for Claude Code, Claude Desktop, Cursor and Windsurf, and the library's client module lets a Browser Use agent call tools on external MCP servers.
  • ✓MIT-licensed open-source library with 117k GitHub stars and BYOK support for OpenAI, Anthropic, Gemini, Groq, OpenRouter, Azure, Bedrock, DeepSeek and local models via Ollama. The cloud needs no subscription and its credits never expire.

Limitations

  • –Complex multi-step workflows on JavaScript-heavy or bot-detection sites can fail, as the LLM reasoning approach adds latency and cost per task compared to deterministic selector-based automation, and success rates on adversarial sites vary significantly.
  • –Open-source self-hosting requires managing your own compute, LLM API keys, and browser infrastructure, meaning teams without DevOps resources face meaningful operational overhead compared to fully managed browser automation alternatives.
  • –The cloud platform privacy policy explicitly states user inputs are used to train Browser Use AI models with no opt-out on standard plans: teams with data sensitivity requirements must use the self-hosted library rather than the cloud API.

Technical Details

Deployment
cloudself-hostedapi
Autonomous rateAutonomously navigates websites, fills forms, clicks elements, extracts data, and completes multi-step browser workflows without per-step human approval once given a natural language task description.
Integrations
OpenAIAnthropicGoogle GeminiGroqOllamaLangChainn8nClaude DesktopCursorClaude CodeWindsurfPlaywrightPuppeteerSeleniumHermes Agent
Security
SOC 2 Type II

Similar agents

Rating

4.7/ 5
★★★★★

Editorial score

How we score this →

Recognition

Rated 4.7 ★Category Leader · Q3 2026MCP Server VerifiedListed 2026
Embed these on your site →

Built for

SaaSEnterpriseB2BDevToolsOpen Source

Leave a review

Never displayed publicly.

Agent Stacks

See workflow stacks that feature Browser Use.

Compare

Related Content

Is this your tool?

Claim this listing free to get your embeddable badges, a review collection link, and the Claimed badge on your listing.

Claim listing →

Free · Every Two Weeks

AI Agent Price & Rating Tracker

Price changes, new agent launches, acquisitions, and rating updates across the AI agent market. Verified against live vendor data, not vendor marketing.

No spam. Unsubscribe anytime. We never share your email.