Glance

Glance is an open-source MCP server that equips AI coding agents like Claude Code with full browser vision and control. It uses Playwright under the hood to let agents navigate pages, click elements, take screenshots, and run multi-step test scenarios — all returned inline so Claude literally sees what the browser renders. With 30 MCP tools, 12 assertion types, and security profiles built in, Glance bridges the gap between AI agents and real-world web app testing.

• Inline Screenshots — Screenshots are returned as base64 images directly in Claude's context, so the agent sees actual rendered pixels rather than guessing at UI state.
• Accessibility Snapshots — Full page structure is returned as a structured text tree with every link, button, input, and heading including labels, roles, and states.
• 30 MCP Tools — Covers browser control (navigate, click, type, hover, drag, scroll, evaluate), test automation (scenario_run, fill_form, auth_flow), and session/visual management.
• Test Scenario Runner — Define multi-step tests in JSON or plain English with 12 assertion types including textContains, isVisible, urlEquals, and consoleNoErrors.
• Visual Regression — Pixel-level screenshot comparison powered by pixelmatch; save baselines and detect visual changes across deployments.
• Network & Console Monitoring — Capture all HTTP requests and console messages to catch failed API calls, CSP errors, and JS exceptions during test runs.
• Security Profiles — Three built-in profiles (local-dev, restricted, open) with URL allowlist/denylist, rate limiting, and JS execution policies.
• Easy Installation — Add to Claude Code with a single claudemcp add glance command, or configure via .mcp.json for Cursor, Windsurf, and Claude Desktop.
• Headed Mode — Set BROWSER_HEADLESS=false to watch the browser operate in real-time as the agent runs tests.
• MCP-Compatible — Works with any MCP-compatible client including Claude Code, Cursor, Windsurf, and Claude Desktop.