</>MCP Agents Market
Skill

Webwright

by microsoft5.9kPythonUpdated 2026-08-03

A simple SWE style browser agent framework that achieves SOTA results on long horizon web tasks.

Claude CodeCodexOpenClawHermes Agent

Webwright is an agent skill framework that transforms coding models into state-of-the-art browser agents for completing complex web tasks. Instead of predicting individual browser actions, it enables AI agents to write complete, reusable Playwright scripts that can spawn browser sessions, inspect pages, and handle long-horizon workflows. The framework integrates as a plugin for Claude Code, Codex, OpenClaw, and Hermes Agent, allowing these AI assistants to automate web interactions through code rather than step-by-step commands. With approximately 1,500 lines of core code, Webwright achieves benchmark-leading results on Online-Mind2Web (86.7%) and Odysseys (60.1%) by treating the browser as a disposable environment rather than stateful workspace.

Key Features

Code-as-action paradigm where agents write full Playwright scripts instead of predicting individual clicks or keystrokes
Plugin manifests for Claude Code, Codex, OpenClaw, and Hermes Agent with shared skill definitions across all platforms
Skill Factory that distills solved tasks into verified, parameterized, reusable code skills that run standalone without model calls
Workspace-as-state design where code, logs, and screenshots persist while browser sessions are ephemeral and disposable
Support for OpenAI, Anthropic, and OpenRouter backends with minimal dependencies (Playwright, httpx, pydantic, typer)
Task Showcase mode that renders completed task results into HTML dashboards with structured JSON reports
Trajectory viewer for comparing token usage and execution paths between different agent harnesses
Slash commands (/webwright:run for one-shot scripts, /webwright:craft for parameterized CLI tools) in supported hosts

Use Cases

  • 01Automating flight searches and price comparisons across travel booking sites with dynamic date ranges
  • 02Scraping structured data from inventory listings, job boards, or real estate sites through reusable scripts
  • 03Building parameterized web automation tools that can be rerun with different arguments without model inference
  • 04Completing long-horizon web workflows like multi-step form submissions, cart checkouts, or account setup flows
  • 05Distilling solved browser tasks into verified skill libraries that agents can reuse across similar tasks
  • 06Benchmarking browser agent performance on complex web navigation tasks with minimal framework overhead

Related Skills

View more

Webwright — FAQ

What is Webwright and how does it differ from other browser agents?+

Webwright is an agent skill that enables coding models to complete web tasks by writing full Playwright scripts instead of predicting individual browser actions. It treats the browser as a disposable environment while persisting code and logs in the workspace, allowing agents to develop, debug, and reuse automation scripts rather than re-discovering interactions from scratch.

How do I install Webwright as a Claude Code plugin?+

In Claude Code, run '/plugin marketplace add microsoft/Webwright' then '/plugin install webwright@webwright'. After installation, start a new Claude Code session so the plugin loads. You can then invoke it with '/webwright:run' or by asking Claude Code to complete web tasks in natural language.

Which AI clients and coding agents work with Webwright?+

Webwright provides plugin manifests and skill definitions for Claude Code, OpenAI Codex, OpenClaw, and Hermes Agent. The same skills/webwright/ folder loads across all four platforms. For standalone use, it can run directly via command-line with Python 3.10+ and Playwright.

What API keys or prerequisites does Webwright require?+

You need Python 3.10+, Playwright with Chromium installed, and an API key for your chosen backend (OPENAI_API_KEY, ANTHROPIC_API_KEY, or OpenRouter credentials). When used as a plugin inside Claude Code or Codex, no additional API key is required beyond your host agent subscription.

Is Webwright free and open source?+

Yes, Webwright is open source and available on GitHub under the microsoft/Webwright repository. The framework itself is free to use, though you'll incur costs from your chosen LLM backend (OpenAI, Anthropic, or OpenRouter) when running standalone, or from your host agent subscription when used as a plugin.

How does the Skill Factory feature work?+

Skill Factory distills solved browser tasks into verified, parameterized code skills that can run standalone without a model. It groups solves of the same template, builds a reusable skill, and replays it to verify correctness before adding it to the library. On WebArena benchmarks, reusing skills lifts accuracy from 55% to 70% while reducing steps and token usage.

How do I install Webwright?+

Open the source repository on GitHub and follow its README. Webwright is a skill — MCP Agents Market links you directly to the official repo.

Is Webwright free?+

Webwright is an open-source project hosted on GitHub. Check the repository for its license and any usage requirements.

Related searches