Browser Use: handing a web browser to an AI agent
Browser Use is an open-source Python library that lets a language model drive a real browser: clicking, filling in forms, reading pages and chaining the steps of a task. The company also sells a pay-as-you-go cloud offering with hosted browsers and agents.
Data as of October 22, 2026
Key figures
| Developer | Browser Use |
|---|---|
| Type | Code |
| Languages | Python |
| Open source | Yes |
| License | MIT |
| First release | November 2024 |
| Suggested level | Intermediate |
| Use cases | web browsing, automation, data |
| Self-hostable | Yes |
| Business model | Freemium |
| Cloud offering | Yes |
| Multi-agent | No |
| MCP support | Yes |
| Bring your own model | Yes |
| Official site | Official site · Documentation · Repository |
GitHub
| Repository | browser-use/browser-use |
|---|---|
| GitHub stars | 118,856 |
| Forks | 13,199 |
| Latest code activity | October 20, 2026 |
| Latest release | 0.13.13: OpenAI browser tools and Cloud fixes (October 15, 2026) |
GitHub activity checked on October 22, 2026.
An agent that uses the web the way you do
The idea behind Browser Use is simple: you describe a task in plain language, such as finding a free slot and booking an appointment, and an AI agent opens a Chromium browser, reads the page, decides where to click or what to type, then repeats until it reaches the goal. It is a form of computer use limited to the browser. The project also serves as a building block for other tools: its website states that Manus built its web automation on the library.
Three ways to use it
The repository lays out three paths. The first is the fully hosted cloud: you send a goal through an API and an agent carries it out on remote browsers, with profiles, recordings and CAPTCHA solving. The second is a command-line interface, paired with a “skill” you install in a coding agent (Claude Code, Codex and others) so it can drive your own Chrome. The third is the Python library, which you run locally from your own code, with the model of your choice and a local or remote browser.
Getting started
The library requires Python 3.11 or later. In a project managed with uv, uv add browser-use installs it; then you add an API key to a .env file and write a short script that creates an agent with a task and a model. New cloud accounts get one dollar of credit to try the hosted version.
Strengths, limits and alternatives
Strengths. MIT license, a large community, model independence, a choice between local and cloud execution, and integration with coding agents and MCP.
Limits. An agent that reads web pages is exposed to prompt injection: hidden text can try to hijack it, so limit the accounts and data it can touch. Every step consumes model calls, which affects cost and speed. Automation can also run into the terms of service of the sites involved. Finally, Browser Use Cloud is hosted only.
Alternatives. OpenHands gives a software development agent a browser among other tools, and smolagents lets you assemble a lightweight Python agent to which you add your own web tools.
Official site: browser-use.com.
Frameworks of the same type
- CAMEL · multi-agent, data, automation
- smolagents · automation, data, multi-agent
- Amazon Bedrock Agents / AgentCore · multi-agent, RAG, workflows
- Atomic Agents · workflows, chatbot, automation
- Claude Agent SDK · coding, automation, workflows
- CrewAI · multi-agent, automation, workflows
Frequently asked questions
Is Browser Use free?
The library is open source under the MIT license and free. You do pay for the language model you use. Browser Use Cloud is pay as you go with no subscription: as of October 22, 2026, 0.02 dollars per browser-hour, 5 dollars per gigabyte of residential proxy traffic, and hosted agents at model cost plus 20%.
Which models can I use?
Browser Use works with several providers, including OpenAI and Anthropic, as well as an in-house model, BU2, available with a Browser Use API key. The quickstart uses an OpenAI key by default.
Is the npm package “browser-use” official?
No. The official library is the Python package from the browser-use/browser-use repository. The npm package with the same name is an unofficial port; check where it comes from before installing it.
Does Browser Use support MCP?
Yes, as an MCP server. The cloud version offers an MCP server for running browsing tasks from clients such as Claude Code, Cursor or Claude Desktop, and the open-source library can itself run as a local MCP server.
Data sources
- github.com
- browser-use.com
- browser-use.com
- docs.browser-use.com
- pypi.org
- GitHub — Repository metadata: GitHub
Data as of October 22, 2026
Report an error
Does a figure, a date or a fact look wrong? Tell us: every report is checked.