AgentIndex icon
AgentIndex
ToolsMCP ServersHOTCategoriesTrendingNewBest ListsCompareGuides
Submit Tool
ToolsMCP ServersHOTCategoriesTrendingNewBest ListsCompareGuides
Home/
Browser Automation/
browser-harness
browser-harness logo

browser-harness

Active·★ 19.1k·MIT·Updated 2026-11-17
★ Most Popular

Browser Harness connects coding agents to Chrome through the DevTools Protocol and exposes browser operations through a CLI or MCP. Its typical use is work that benefits from an existing browser session, while its local workspace lets an agent retain task-specific helper code. That retained code should be reviewed as executable automation; it does not establish that a model has trained itself or learned every similar website.

We installed browser-harness 0.1.13 with its MCP extra and tested an isolated headless Chrome 156 profile against a synthetic local page. Discovery returned 23 tools. Input, coordinate clicking, a normal file upload, localStorage after reload, switching between owned tabs and screenshot capture succeeded. The saved result text and visible filename were checked as outcomes rather than assuming that a returned tool call meant the action succeeded.

A missing-selector call returned an error field inside JSON content while the outer MCP isError flag remained false; the next page inspection still worked. Hosts therefore need to parse the tool's content and define their own completion checks. This test did not cover personal logins, authorization dialogs, real-site tasks, autonomous model planning or comparative performance. Our loopback proxy setup also needed adjustment before the CDP connection worked.

Read the full review, verification scope and sources →

Hands-on tested · Tested on 2026-11-04 with browser-harness 0.1.13 (MCP controls only).

Setup: Python 3.12.13, Node.js 26.6.0, Chrome 156.0.8078.80; new headless profile and synthetic localhost page

Commands run

python3 doc/review-fixtures/browser-harness/run_lab.py --harness-mcp tmp/deep-reviews-20261104-batch2/browser-venv/bin/browser-harness-mcp --test both

What we saw

  • Discovered 23 tools; form entry, coordinate click, file input, localStorage reload, tab switching, missing-selector recovery and screenshot checks passed.
  • The missing selector produced JSON error content with outer isError false; the next page read succeeded. Screenshot was 756 × 494.
  • The loopback connection needed a proxy bypass. No personal login, permission-dialog, real-site, model-planning or cross-tool performance test.

Data as of 2026-11-12.

License
MIT
Latest release
v0.1.13 · 2026-09-04
Last commit
2026-11-11
Contributors
104
Forks
1,884
Open issues and PRs
459
First published
2026-04-17

Source: GitHub, npm and PyPI public APIs.

#agent-framework#ai-agent#browser-agent#browser-automation#browser-use#browser-use-box#browser-use-cloud#cdp
$ Install
$ uv tool install --python 3.12 'browser-harness[mcp]==0.1.13'
↗ Visit site★ GitHub
01

Features

01Drives your running Chrome, logged in as you, over the DevTools Protocol
02Agents write missing helpers into a local workspace while the core source stays protected
03Set up by pasting one prompt into Claude Code or Codex
04Any MCP client can connect through browser-harness-mcp
02

Why choose it

+The controlled MCP workflow completed input, clicking, file selection, tab switching and screenshots in a disposable Chrome profile.
+An explicit CDP endpoint allowed the review to isolate browser control from personal sessions and model behavior.
+Task helpers can persist in a local workspace for later review and reuse instead of changing the core package each time.
03

Trade-offs

!The default browser-instance connection requires deliberate profile and account selection; it is not a per-business-action permission boundary.
!Some errors are JSON content with isError false, so integrations need content-aware outcome handling.
!Connection setup and page state affect reliability; this fixture did not verify real logins or long autonomous tasks.
05

Maintenance check

ActiveChecked: 2026-11-10
Last commit
2026-11-09
Commits, last 90 days
100+
Releases, last 90 days
5
Latest release
v0.1.13 (2026-09-04)
Open issues
454
Recent issues with at least one comment
0/5
Commits per week, last 12 weeks
Commits per week, last 12 weeks: 16, 48, 2, 2, 5, 4, 56, 3, 19, 2, 12, 5

Oldest week on the left; the latest four weeks are orange. Source: GitHub commit activity.

Status rule: archived on GitHub = Archived; last commit within 90 days = Active; 91 to 365 days = Slowing; older = Stale. All figures come from the GitHub API on the date shown.

04

Who it fits

Pick it if
  • A coding agent needs browser actions in a deliberately chosen session or profile.
  • You can review reusable helpers and implement visible outcome checks and error recovery.
Look elsewhere if
  • Your main requirement is deterministic CI testing of your own application with a clean browser.
  • You expect the browser connection itself to enforce granular business permissions.
06

Compatibility

Claude Code
Native
Verified via docs
Codex
Native
Verified via docs
Cursor
Via MCP
Verified via docs
Chrome
Browser
Verified via docs
07

Quick start

1
$ uv tool install --python 3.12 'browser-harness[mcp]==0.1.13'
08

Use cases

↳Have an agent complete tasks on sites where you are already logged in
↳Download your own content in bulk, such as your latest video posts on X
↳Give a coding agent browser actions such as uploading files
09

Alternatives

⚡Top VS:
VSGenericAgent★14.4k→VSJobs_Applier_AI_Agent_AIHawk★32.6k→VScdpilot★33→VSwebpage-mcp★29→
GenericAgent logo
GenericAgent★ 14.4k
Self-evolving agent: grows skill tree from 3.3K-line seed, achieving full system control with 6x less token consumption
VSCompare→
open-codex-browser-use logo
open-codex-browser-use★ 425
👾 Open Browser Use – Open-Source Alternative to Codex Browser Use
Jobs_Applier_AI_Agent_AIHawk logo
Jobs_Applier_AI_Agent_AIHawk★ 32.6k
Give it your resume, get the jobs that fit, select them and apply. Runs in your terminal, with a browser you can watch.
VSCompare→
wmux logo
wmux★ 456
AI Agent Terminal for Windows — Run Claude Code, Codex, Gemini CLI side by side with split terminals, browser automation via CDP, MCP integration, session persistence, and smart notifications
cdpilot logo
cdpilot★ 33
Zero-dependency browser automation CLI. 70+ commands, 10 test assertions, smart commands (click/fill by text — no LLM needed). MCP server for AI agents with 500x fewer tokens. Extract, observe, script runner. 50KB, pure CDP.
VSCompare→
webpage-mcp logo
webpage-mcp★ 29
Turn your existing webpages into an MCP server for agent control
VSCompare→
hermes-browser-extension logo
hermes-browser-extension★ 1.9k
Browser-native side panel for Hermes Agent — connect web context to your local Hermes runtime.
VSCompare→
hermes-agent logo
hermes-agent★ 264.2k
The agent that grows with you
VSCompare→
See all alternatives →

Comments

Log in to leave a comment

No comments yet. Be the first!

On this page
01Features02Why choose it03Trade-offs05Maintenance check04Who it fits06Compatibility07Quick start08Use cases09Alternatives
browser-use /
browser-harness
↗

Browser controls for coding agents with reusable local helpers

19.1k1.9kMIT
Last commit1d ago
StatusActive
CategoryBrowser Automation
Contributors104
Latest releasev0.1.13
Links
Documentation↗Discussion↗Issues↗Releases↗
Cloud Infrastructure
Ad

Deploy on DigitalOcean

Host your MCP servers, Docker containers, and AI agents with $200 free credit over 60 days.

Claim $200 Free Credit
© 2026 AgentIndex.app|Built by a 10-year iOS Developer.
Best by use caseAboutContactAdvertisePrivacy PolicyTerms of UseQYSGitHubBuy me a coffee ☕

Browse by Category

MCP ServersCode AssistantWorkflow AutomationRAG / Knowledge BaseMulti-AgentBrowser AutomationLLM InfraDev ToolingObservability

Not affiliated with Anthropic, OpenAI or Microsoft.