Mcp Server Browser
Give your agent real browser control for navigation, UI checks, and web research through MCP instead of manual screenshots.
Overview
io.github.bytedance/mcp-server-browser is a MCP server for the Build phase that lets agents drive Chrome, Edge, or Firefox for web access and automation.
What is this MCP server?
- npm package @agent-infra/mcp-server-browser with stdio transport
- Browser channel argument: chrome, edge, or firefox
- Optional CDP endpoint, WebSocket endpoint, and custom executable_path
- output-dir for artifacts such as captures from automated runs
- Proxy bypass and related launch flags for controlled environments
- Server version 1.0.0
- 3 browser channel values: chrome, edge, firefox
- npm identifier @agent-infra/mcp-server-browser
Community signal: 36.3k GitHub stars.
What problem does it solve?
Your agent cannot see or interact with live web UIs, so you keep pasting HTML snippets and manual test steps that drift from reality.
Who is it for?
Builders shipping web SaaS or extensions who need the agent to verify flows, scrape public pages, or debug UI behavior in a real browser.
Skip if: Unattended production crawlers, compliance-heavy audits, or teams that forbid any browser automation from dev machines.
What do I get? / Deliverables
After registration, the agent can open pages and run browser-backed workflows through MCP using your chosen browser channel and CDP or WebSocket endpoints.
- Agent-controlled browser sessions on chrome, edge, or firefox channels
- Configurable CDP or WebSocket attachment for existing browser instances
- Captured outputs in a chosen output directory when runs produce files
Recommended MCP Servers
Journey fit
Browser automation is first shelved under Build integrations because agents most often need it to exercise web apps and external sites while coding. It connects the agent runtime to Chrome, Edge, or Firefox via CDP or WebSocket endpoints—classic integration work, not pure frontend styling.
How it compares
MCP browser driver, not a full E2E test framework or SEO crawler skill.
Common Questions / FAQ
Who is io.github.bytedance/mcp-server-browser for?
Solo builders and small teams who want MCP-driven browser access from Claude Code, Cursor, or similar agents while building or checking web products.
When should I use io.github.bytedance/mcp-server-browser?
Use it when you need the agent to load real pages, exercise UI flows, or research sites during build, test, or early discovery work.
How do I add io.github.bytedance/mcp-server-browser to my agent?
Add @agent-infra/mcp-server-browser to MCP config over stdio, set the browser argument to chrome, edge, or firefox, and optionally pass cdp-endpoint, ws-endpoint, or executable_path.