Dev Browser is a Claude Code plugin designed to equip AI agents with web browsing capabilities, enabling them to interact with and test live applications directly within a development environment. It targets developers using Claude Code who need to automate browser interactions for tasks like form submission, button clicking, and verification, thereby streamlining development and debugging workflows.

How It Works

Dev Browser employs a stateful server architecture combined with agentic script execution, offering a hybrid approach that balances speed and robustness. Unlike simpler observe-think-act loops (Playwright MCP) that are slow and context-heavy, or fragile end-to-end scripts (Playwright Skill) that restart from scratch, Dev Browser maintains a persistent Playwright server. This statefulness allows pages to remain alive across multiple interactions, enabling flexible execution where agents can use full scripts when a plan is clear or fall back to step-by-step exploration. It also provides LLM-friendly DOM snapshots and can leverage codebase awareness to inform debugging.

Quick Start & Requirements

• Prerequisites: Claude Code CLI, Bun runtime (v1.0 or later).
• Installation (Claude Code):
  1. Add marketplace: /plugin marketplace add sawyerhood/dev-browser
  2. Install plugin: /plugin install dev-browser@sawyerhood/dev-browser
  3. Restart Claude Code.
• Installation (Amp):
  1. Create skills directory: mkdir -p ~/.claude/skills
  2. Clone and copy skill: git clone https://github.com/sawyerhood/dev-browser /tmp/dev-browser-skill && cp -r /tmp/dev-browser-skill/skills/dev-browser ~/.claude/skills/dev-browser && rm -rf /tmp/dev-browser-skill
  3. Start server in a separate terminal: cd ~/.claude/skills/dev-browser && bun install && bun run start-server (Keep this terminal running).
• Links: GitHub Repository (implied by install commands)

Highlighted Details

• 14% faster and 39% fewer turns than Playwright MCP.
• 52% faster and 24% fewer turns than Playwright Skill.
• 39% cheaper than both alternatives based on benchmark costs.
• Achieved 100% success rate in benchmarks across 3 runs.
• Provides LLM-friendly structured DOM snapshots for AI understanding.

Maintenance & Community

The project is authored by Sawyer Hood. No specific community links (Discord, Slack) or detailed roadmap information are present in the provided README.

Licensing & Compatibility

• License: MIT.
• Compatibility: The MIT license generally permits commercial use and integration into closed-source projects, with attribution.

Limitations & Caveats

When using with Amp, users must manually start the dev-browser server in a separate terminal and keep it running, as Amp cannot manage background processes for this skill.