Browserr — headless browser engine for AI agents

Scope

01 / 05

Everything an agent needs.

Built agent-first: extraction over pixels, selectors over coordinates, failures you can catch instead of hangs. The automation surface is deliberately small.

Command	Covers
read	Structured page reads. Prose via read_page(), or read_structure(): headings, buttons, inputs, selects, links, images — each with a viewport rect and a CSS selector that feeds the element commands.
act	Selector-based interaction. click_element() scrolls into view and clicks with a real pointer event; set_value() uses the native setter plus input/change events, so React and Vue controlled inputs accept it.
wait	SPA-aware timing. wait_for_selector() polls until an element is visible; wait_ready() gates on document.readyState. No sleeps, no flaky scripts.
observe	Console capture. Page console.log/warn/error arrives as events — and Servo routes network and TLS load errors through the same channel, so agents see why a page failed.
files	Downloads. download(url, path) fetches straight to disk with size caps and an ok/bytes/error event. Public URLs only — no page cookies involved.
survive	Fast failure, not hangs. EngineDeadError surfaces a dead engine in 0.07 s instead of a timeout. ping health-checks. Ephemeral profiles give every run a clean slate.
block	Adblock built in. EasyList-based subresource blocking at the engine's request-interception hook, auto-refreshed weekly. Never blocks main-frame documents.
gpu	Fast rendering. Servo renders on the GPU; frames cross processes as D3D12 shared textures (no pixel copies) with a byte-frame fallback. Screenshots on demand, per tab.
secure	Token-gated attach. --driver-port accepts multiple concurrent drivers; --driver-token requires an auth line first. Unauthenticated sockets get nothing — not even events.

Protocol

02 / 05

The entire protocol, in full.

Newline-delimited JSON over loopback. One client, no hidden state — the opposite of CDP's 1000+ endpoints. Events: url, title, history, cursor, favicon, pwcapture, evalresult, pong, console, download.

Commands: 15
MCP tools: 19
Binary: ~105 MB
CDP endpoints: 0
Dependencies: 0

SDK

03 / 05

Three ways in.

Same engine underneath; pick the integration you already have.

Surface	Covers
mcp	Model Context Protocol server with 19 tools, stdlib only. Any MCP-capable agent — Claude, Kai, your own — drives Browserr with zero integration code.
python	driver.py wraps the raw protocol: navigation, extraction, element commands, console capture, downloads, screenshots, a PNG helper. No dependencies.
raw	Anything that can write JSON lines to a socket. The whole protocol fits on one screen; bindings in any language are an afternoon.

Position

04 / 05

Where it sits in 2026.

Chromium wrappers are the default; alternative engines are the bet. This is the honest comparison.

	Browserr	Lightpanda	Obscura	Chromium headless
engine	Servo	own (Zig)	own (Rust)	Blink
js engine	SpiderMonkey	own	own	V8
mcp-native	yes, 19 tools	no	partial	via wrappers
protocol size	15 commands	CDP-compatible	own + CDP	1000+ endpoints
footprint	one ~105 MB binary	~16× less than Chromium	~30 MB	~200 MB+
web compat	Servo 0.8 (text, docs, forms)	growing	growing	full

Honest limits. No site sandbox on Windows (treat page loads as untrusted; use ephemeral profiles), web compatibility bounded by Servo 0.8, Windows x64 only, and rendering needs a GPU with a D3D11 driver. Text, documents, search and forms work well; heavy SPAs and DRM video may not.

Install

05 / 05

Get Browserr in seconds.

One zip: the engine binary, the Python SDK, and the MCP server. No account, no telemetry.

Windows x64

Download the zip, unpack anywhere, and point your agent at the MCP server — or drive the engine directly from Python. A GPU with a D3D11 driver is required for rendering.

↓ Download for Windows

Format .zip Contents engine.exe · driver.py · browserr_mcp.py Requires GPU (D3D11) · Python 3.10+

Start

Start driving the web with your agents.

Free, open source, no account required.

↓ Download for free Star on GitHub