harness-browser 0.1.3

AI-friendly browser automation via CDP with profile-based login persistence

Harness Browser

AI-friendly browser automation via Chrome DevTools Protocol (CDP).

English · 中文

---

Highlights

• Pure CDP, Zero Playwright — Direct WebSocket connection to Chrome DevTools, no browser binaries shipped, no driver layer
• Token-Efficient DOM — 4-level output from ~50 to ~3000 tokens; interactive mode returns only clickable/typeable elements with stable refs
• Login Persistence — One Chrome user-data-dir per profile; log in once, every subsequent run reuses cookies and storage
• Agent-First API — Stateless browser_tool() function, MCP server, and Claude Code skill ready to drop into any agent project
• Full Observability — Per-action metrics (duration_ms, estimated_tokens, screenshot_size_kb) plus lifecycle hooks

---

Overview

Harness Browser is an agent-first browser runtime built on pure CDP. It provides predictable DOM snapshots, persistent profile sessions, and a typed Python API designed for LLM tool-calling. No Playwright, no Selenium — just a direct WebSocket connection to Chrome.

The library solves three problems for AI agents:

1. Token cost — Multi-level DOM output keeps context windows lean
2. Element stability — Ref-based targeting survives layout reflows
3. Authentication — Profile persistence eliminates repeated logins

---

Core Technology

Component	Technology	Purpose
Transport	WebSocket (websockets>=12.0)	Direct CDP communication with Chrome
DOM Engine	Custom tree walker	Multi-level DOM serialization with ref assignment
Configuration	Pydantic models	Typed settings with env-var override
MCP Server	mcp>=1.0	Expose browser actions as MCP tools
Profiles	Chrome user-data-dir	Persistent login state per named profile

---

Features

• Pure CDP — Direct WebSocket connection, no Playwright dependency
• Profile-based login persistence — Chrome user-data-dir per profile, cookies/sessions reused across runs
• 4-level DOM output — minimal (~50 tokens), interactive (~200–500 tokens), full (~1000–3000 tokens), structured (JSON)
• Ref system — Stable element references across actions, invalidated on navigation
• Hook system — before_action, after_action, action_error, page_navigated
• Per-action metrics — duration_ms, estimated_tokens, screenshot_size_kb
• Environment-variable configuration — All paths, ports, and timeouts configurable without code changes
• Remote/Docker Chrome support — Bypass launcher via BROWSER_USE_CDP_WS_URL
• MCP Server — Expose actions as MCP tools for Claude Code and other MCP clients
• Drop-in Claude Code skill — Copy skills/harness-browser/ into any agent project
• Strict typing — mypy strict, ruff clean, comprehensive test coverage

---

Quick Start

Requirements

• Python 3.11+
• Chrome or Chromium

# Ubuntu/Debian
sudo apt install chromium-browser

# macOS
brew install --cask google-chrome

Installation

pip install harness-browser

# Optional: download a Playwright-managed Chromium into the standard
# cache (~/.cache/ms-playwright/...) when you don't have a system Chrome.
# This installs `playwright` itself if it isn't already present, then
# fetches the browser binary. Set HARNESS_SKIP_PLAYWRIGHT_PIP=1 to skip
# the pip step (e.g. in pre-baked images).
harness-browser install-browser

Python API

import asyncio
from harness_browser import BrowserSession

async def main():
    async with await BrowserSession.create(profile="default") as sess:
        await sess.navigate("https://example.com")
        result = await sess.dom_tree(level="interactive")
        print(result.content)
        # → [ref=inp_1] input[text] placeholder="Search"
        # → [ref=btn_2] button "Go"
        await sess.click(ref="btn_2")

asyncio.run(main())

Stateless Tool Interface (for AI Frameworks)

from harness_browser import browser_tool

# All calls route to the same session by profile name
result = await browser_tool(action="navigate", url="https://github.com", profile="work")
result = await browser_tool(action="dom_tree", level="interactive", profile="work")
result = await browser_tool(action="click", ref="btn_search", profile="work")
result = await browser_tool(action="type", text="harness", profile="work")

CLI

Every action is also a shell command. Sequential calls on the same --profile attach to the same Chrome process, so refs and login state carry across invocations:

# Drive the browser
harness-browser navigate "https://example.com" --profile work
harness-browser dom-tree --profile work
# → [ref=inp_1] input[text] placeholder="Search"
# → [ref=btn_2] button "Go"
harness-browser click --ref inp_1 --profile work
harness-browser type "harness" --profile work
harness-browser click --ref btn_2 --profile work
harness-browser screenshot --path /tmp/result.png --profile work

# Tear down (Chrome itself stays running for later attach)
harness-browser close-session --profile work

Add --json to any command for the full structured ToolResult (useful in scripts), and --auto / --headed / --headless to override the launch mode on the first call per profile. Run harness-browser --help for the full list of subcommands.

MCP Server

python -m harness_browser.mcp_server

Add to Claude Code settings.json:

{
  "mcpServers": {
    "harness-browser": {
      "command": "python",
      "args": ["-m", "harness_browser.mcp_server"],
      "env": {
        "BROWSER_USE_MODE": "auto",
        "BROWSER_USE_PROFILES_DIR": "/data/browser-profiles"
      }
    }
  }
}

Available MCP tools: browser_navigate, browser_dom_tree, browser_screenshot, browser_click, browser_type, browser_eval_js, install_browser.

---

DOM Levels

Level	Tokens	Use Case
minimal	~50	Confirm page loaded, check title/URL
interactive	~200–500	Find clickable/typeable elements (default)
full	~1000–3000	Read page content
structured	varies	JSON for programmatic processing

---

Login State Reuse

Profiles persist Chrome sessions in ~/.harness-browser/profiles/<name>/:

# First run: navigate to login page, log in manually
await browser_tool(action="navigate", url="https://github.com/login", profile="github")

# All future runs: login state reused automatically
await browser_tool(action="navigate", url="https://github.com/settings", profile="github")

---

Hook System

async with await BrowserSession.create(profile="work") as sess:
    @sess.on("before_action")
    async def log_action(event):
        print(f"[{event['action']}] starting")

    @sess.on("after_action")
    async def log_metrics(metrics):
        print(f"  done in {metrics.duration_ms}ms (~{metrics.estimated_tokens} tokens)")

    await sess.navigate("https://example.com")

---

Screenshots

screenshot writes a PNG to disk and returns its path — never raw base64. That keeps token usage flat regardless of image size.

# Default: timestamped file in BROWSER_USE_SCREENSHOTS_DIR
result = await sess.screenshot()
# → /home/user/.harness-browser/screenshots/harness-1779462725763.png

# Full scrollable page
await sess.screenshot(full_page=True)

# Crop to a single element
await sess.screenshot(element_ref="btn_2")

# Pin the file path
await sess.screenshot(path="/tmp/latest.png")

---

Configuration

All settings can be configured via environment variables — no code changes required.

Environment Variable	Default	Description
BROWSER_USE_PROFILES_DIR	~/.harness-browser/profiles	Root directory for Chrome user-data-dirs
BROWSER_USE_SCREENSHOTS_DIR	~/.harness-browser/screenshots	Directory for PNG screenshots
BROWSER_USE_CDP_HOST	localhost	Host serving Chrome's CDP endpoint
BROWSER_USE_CDP_PORT_START	9222	First CDP debug port
BROWSER_USE_MODE	auto	Launch mode: auto / headed / headless
BROWSER_USE_CHROME_BIN	auto-detect	Absolute path to Chrome/Chromium
BROWSER_USE_CDP_TIMEOUT	30.0	CDP command timeout (seconds)
BROWSER_USE_CDP_WS_URL	—	Direct WebSocket URL (bypasses launcher)

---

Actions Reference

Action	Required	Optional
navigate	url
dom_tree		level (default: interactive)
screenshot		element_ref, full_page, path
click	one of: ref, selector, x+y
type	text	ref
scroll		direction, amount
hover	ref
eval_js	expression
go_back
go_forward
reload
list_tabs
new_tab		url
switch_tab	tab_id
close_tab		tab_id
close_session

---

Development

# Clone
git clone https://github.com/orcakit/harness-browser.git
cd harness-browser

# Install with dev extras
uv sync --extra dev

# Install pre-commit hooks
pre-commit install

# Run tests
make test

# Lint + type check
make lint

# Format
make format

# Build wheel
make build

---

Related Projects

Project	Description
harness-agent	Production-grade AI agent platform built on LangChain Deep Agents
harness-memory	Pluggable memory system with hierarchical recall and FTS
harness-im-bridge	Multi-platform IM channel bridge for AI agents

---

Contributing

See CONTRIBUTING.md.

License

MIT