Browser Harness
Self-healing browser agent harness built on CDP — agents rewrite their own helpers mid-task to finish any job.
How the 6.6 was reached
- Quality of output6/10
18,208 GitHub stars and active pushes indicate solid adoption, but this is well below the 8 anchor of tens of thousands; self-healing claim is vendor-stated. github.com
- Ease of use6/10
Setup by pasting a prompt into Claude Code or Codex with an install.md fallback; no docs page found. github.com
- Pricing value9/10
Fully free under the MIT licence with no pricing tiers, though it needs an agent to run it. opentools.ai
- API and integration quality6/10
MCP server exposes browser helpers to Claude Code, Devin and Cursor; no separate docs or SDK found. github.com
- Solves the problem it claims to solve6/10
Strong GitHub adoption of 18k stars, but no independent evidence that agents self-heal on any task. github.com
6.6/10 is the mean of the 5 criteria that apply. Scored on the five-criteria rubric v1, 29 September 2026. How scoring works
Pricing
- MIT licensed
- Self-hosting on local Chrome
- Full CDP access
- Community support
- Managed stealth browser fleet
- Anti-detection headers
- No local Chrome required
- 24/7 always-on execution
- Dedicated browser sandbox
- Persistent agent sessions
Key Features
- Self-healing agent loop — helpers are rewritten mid-task when they fail
- Direct Chrome DevTools Protocol (CDP) integration for low-latency control
- Local Chrome support for development and testing
- Stealth browser fleet option to bypass bot detection
- 24/7 cloud execution via Browser Use Box
Pros & Cons
Pros
- MIT licensed and open source — no vendor lock-in for the core harness
- Self-healing design cuts brittle automation failures without manual patching
- Thin CDP layer keeps overhead low compared to Playwright or Selenium wrappers
- Flexible deployment: local, stealth cloud, or dedicated box
Cons
- Pricing for stealth browsers and Browser Use Box is opaque — no public tiers
- Requires comfort with LLM agent integration; not a no-code tool
- Very new project with limited community resources and production track record
- CDP-only means no Firefox or Safari support
Browser Harness is a sharp, developer-focused primitive for building resilient browser agents — the self-healing loop that rewrites helpers mid-task is a genuinely useful idea for flaky automation. It sits one abstraction below browser-use and Playwright, which suits teams who want raw CDP control. Too early-stage and thinly documented for teams needing production SLAs today, but worth watching.
Try Browser Harness →Added to scored.tools on
Show this score
Free for any tool, at any score. It shows the live score, so it changes when the score does. Paste this where you want it:
<a href="https://scored.tools/tools/browser-harness/"><img src="https://scored.tools/badge/browser-harness.svg" alt="Browser Harness on scored.tools: 6.6/10" width="160" height="40"></a>Do you run Browser Harness? Claim this page to confirm the facts, correct one or respond. It is free and never changes the score.
Competitors to Browser Harness
Other tools in the automation category worth comparing.
Browser Use
7.8/10Open-source AI-powered browser automation platform for web agents and tasks
Cua
6.8/10Cloud desktops for AI agents with support for any OS including macOS, Windows, Linux, and Android.
Firecrawl
7.4/10Web scraping and search API designed specifically for AI agents and applications.
E2B
7.6/10Secure cloud sandboxes for AI agents to run code and interact with real-world tools safely.
More Articles Featuring Browser Harness
Learn Harness Engineering Review 2026: Free Course For AI Agent Builders
A free, project-based course teaching the environment and verification scaffolding that makes Claude Code and Codex agents actually reliable in production.
reviewBrowser Harness Review 2026: Self-Healing CDP Agent Primitive
Browser Harness is an MIT-licensed CDP agent layer whose helpers rewrite themselves mid-task. Sharp idea, early-stage execution. Honest builder's review.