Glance

Product Introduction

1. Definition: Glance is a high-performance, open-source Model Context Protocol (MCP) server built on the Playwright framework that provides AI agents with a fully functional, controllable Chromium browser. It acts as a specialized bridge between Large Language Models (LLMs) and the web, enabling terminal-based agents like Claude Code, Cursor, and Windsurf to execute complex browser-based tasks.

2. Core Value Proposition: Glance exists to solve the "context gap" in AI agentic workflows by providing "eyes" to text-based models. By delivering inline screenshots, accessibility snapshots, and 30 dedicated automation tools, it allows AI agents to navigate, interact with, and test web applications with a 97% success rate. Key value drivers include automated E2E testing, visual regression monitoring, and secure browser automation directly from a developer's CLI or IDE.

Main Features

1. 30 Integrated MCP Tools: Glance provides a comprehensive suite of tools categorized into Browser Control (19 tools), Test Automation (7 tools), and Session Management (4 tools). This includes granular actions such as hover, drag, press_key, evaluate (for custom JavaScript execution), and tab_management. The toolset is designed to mirror human interaction patterns, allowing agents to manipulate the DOM with high precision.

2. Multi-Modal Feedback (Screenshots and Accessibility Snapshots): To ensure the agent "sees" what it is doing, Glance returns high-resolution base64 screenshots directly into the LLM's response stream. Complementing this is the "Accessibility Snapshot" feature, which converts the page structure into a hierarchical text tree, detailing roles, states, and labels. This dual-layered feedback mechanism eliminates LLM "hallucinations" regarding page layouts.

3. Advanced Test Scenario Runner: The platform includes a native E2E test engine featuring 12 assertion types, including textContains, isVisible, countEquals, and consoleNoErrors. It supports multi-step scenario execution with state persistence, allowing users to define complex auth-flow testing and form-filling logic in plain English or JSON.

4. Security Profiles and URL Filtering: Glance incorporates robust security controls to prevent unintended agent behavior. It features three distinct security profiles—local-dev, restricted, and open—and supports URL allowlists/denylists, rate limiting, and strict JavaScript execution policies.

5. Visual Regression Testing: Powered by the pixelmatch library, Glance enables agents to save baseline screenshots and perform pixel-level comparisons. This allows for automated detection of UI shifts or visual bugs during the deployment pipeline.

Problems Solved

1. Pain Point: Agentic Blindness and DOM Complexity. Traditional AI agents often struggle with modern single-page applications (SPAs) or shadow DOMs because they rely on raw HTML which can be too token-heavy or confusing. Glance solves this by providing visual and structural snapshots that prioritize the user-facing interface.

2. Target Audience:

• QA and Automation Engineers: Who need to build and run E2E test suites using natural language via Claude Code.
• Full-Stack Developers: Looking to debug localhost environments or run visual regression tests without leaving their terminal.
• AI Agent Developers: Who require a reliable, secure browser environment to extend the capabilities of their custom agents.

3. Use Cases:

• Automated Login Flow Testing: Navigating through authentication barriers and verifying post-login states.
• Continuous Integration/Deployment (CI/CD): Running headless visual audits of web apps during PR checks.
• Live Web Scraping and Interaction: Accessing dynamic content that requires JavaScript execution and human-like interaction (e.g., clicking "load more" buttons).

Unique Advantages

1. Differentiation: Unlike standard browser automation scripts, Glance is natively built for the Model Context Protocol (MCP). It is optimized for the latency and token constraints of LLMs, providing structured feedback that models can actually parse efficiently. Its "battle-tested" 97% pass rate across 300+ production test steps sets it apart from more experimental browser-agent integrations.

2. Key Innovation: The integration of the "Accessibility Snapshot" as a primary navigational aid is a significant technical shift. By focusing on accessibility roles and states, Glance provides the LLM with the most semantically meaningful data, which is far more reliable for automation than raw CSS selectors or XPaths.

Frequently Asked Questions (FAQ)

1. How do I install Glance for Claude Code? Glance can be added to Claude Code using the command claude mcp add glance -- npx glance-mcp. For those who want to watch the browser in real-time, you can disable headless mode by setting the environment variable BROWSER_HEADLESS=false during the configuration process.

2. Does Glance support visual regression testing and pixel comparisons? Yes. Glance includes a specialized visual_compare tool. It allows users to establish baseline screenshots of their application and then compare subsequent runs to detect visual changes. It uses the pixelmatch library to identify differences at the pixel level, making it ideal for catching UI regressions.

3. Is Glance an open-source tool? Glance is fully open-source and released under the MIT License. It is maintained by DebugBase and welcomes community contributions via its GitHub repository. It can be installed globally via npm using npm install -g glance-mcp.

4. What AI agents are compatible with Glance? Glance is an MCP-compliant server, meaning it works out of the box with any client that supports the Model Context Protocol. This includes Claude Code, Cursor, Windsurf, and the Claude Desktop application.