About BrowserControl¶
BrowserControl is an open-source MCP server that gives AI agents a real browser they can see, click, type, and debug — local, private, zero-cost.
Mission¶
Give AI agents a better way to use a browser — one they can actually see, running on your machine, free for anyone to use.
Agents are asked to work on the web every day, and most of the tooling still hands them a DOM and hopes for the best. BrowserControl exists to close that gap: show the agent the page, let it point at what it wants, and keep the whole loop local, private, and free. Every design decision is measured against that — does it make an agent more capable in a browser, and does it stay out of the user's way?
Origin¶
BrowserControl started as a single-engineer prototype to scratch one itch: browsers shouldn't be hard to drive for AI agents. Existing tools used fragile selectors, required cloud APIs, or cost money per action. None of that made sense for a tool that's running on the user's machine anyway.
The first version shipped ~25 tools organized into navigation, interaction, forms, content, devtools, recording, and tabs. Subsequent releases added Set of Marks (SoM), shadow DOM traversal, native file uploads, and built-in devtools.
Project facts¶
| License | MIT |
| Python | 3.11+ |
| Stack | FastMCP, Playwright, Pillow |
| CI | GitHub Actions — Ubuntu, Windows, macOS |
| First release | June 2025 |
| Current version | 0.1.4 |
| Repository | github.com/adityasasidhar/browsercontrol |
| PyPI | pypi.org/project/browsercontrol |
Maintainers¶
- Aditya Sasidhar — original author and maintainer.
See the CONTRIBUTING.md for how to get involved.