Browserbase MCP Server
https://github.com/browserbase/mcp-server-browserbase- Category
- Developer Tools
- Rank
- No. 2207Tools index
Previous survey · No. 2120 ·
- Pricing
- Open Source
- Type
- TOOL
- Builder
- browserbase
- GitHub
- 3.4k stars
- Latest release
- v3.0.0
- Date
About
MCP server that lets LLMs control a real browser via Browserbase and Stagehand — fully-managed headless infrastructure for agent automation.
What it does
This server gives an MCP client six browser tasks: open or reuse a session, close it, visit pages, perform natural-language actions, inspect actionable elements, and extract page data. Stagehand interprets the instructions while Browserbase supplies the browser session.
Why it's ranked here
The case is mixed. The focused six-tool surface, two transports, packaged command, Docker route, and configurable model support make the implementation useful as a compact reference. However, the repository is archived, explicitly unmaintained, and disclaims representing current production services. The hosted service is the recommended practical choice.
What's good
The interface stays small and understandable while covering session lifecycle, navigation, interaction, observation, and extraction. Users can connect through standard input or streamable HTTP. Self-hosters get command-line controls for proxies, persistent contexts, viewport size, session retention, verified identity, network binding, and model selection.
Tradeoffs
Self-hosting still requires Browserbase credentials and a project identifier, so it does not provide browser execution independent of Browserbase. The default model needs a corresponding API key, while custom models require their provider key and Stagehand support. Verified identity is restricted to Scale Plan users. Most importantly, this code receives no maintenance.
How to use it well
Use it as a reference implementation, or for a controlled MCP workflow that needs natural-language browsing and accepts Browserbase as its session backend. Prefer the hosted endpoint when ease of setup and current service behavior matter. Do not choose this repository for an actively maintained self-hosted browser stack or automation independent of Browserbase.
Technical notes+
package.json defines an ESM TypeScript package published as @browserbasehq/mcp, with a CLI entry through cli.js and a library entry through index.js. src/index.ts registers Zod-validated tools and MCP resource handlers. src/transport.ts implements STDIO and Streamable HTTP transports, keeping HTTP sessions in memory. src/sessionManager.ts creates Stagehand instances against Browserbase, tracks active sessions, guards concurrent default-session creation with a promise, retries creation once, and closes sessions through Stagehand. src/config.test.ts contains Vitest coverage for the verified-identity compatibility alias. Configuration defaults are inconsistent: src/config.ts uses google/gemini-2.5-flash-lite, while fallbacks in src/transport.ts and src/sessionManager.ts use gemini-2.0-flash.
Observed
- License
- Apache 2.0
- Primary language
- TypeScript
- Package
- Public npm package named @browserbasehq/mcp
- Interfaces
- MCP server with STDIO and Streamable HTTP transports, plus CLI and module entry points
- Installation surfaces
- Hosted endpoint, npx package execution, direct source build, and Docker image build
- Maintenance status
- Repository is archived and explicitly described as no longer maintained
- Tests
- Vitest is configured, and the supplied tree includes a configuration test file
Read from README.md, package.json, cli.js, index.js, src/index.ts, src/server.ts, src/config.ts, src/context.ts, src/program.ts, src/transport.ts, src/config.test.ts, src/sessionManager.ts, src/tools/act.ts, src/tools/tool.ts, src/tools/index.ts.
What it can do
Navigate to web pages
URL or web address → Loaded web page content
Click on web elements
Element selector or coordinates → Element interaction result
Fill out web forms
Form field data and selectors → Completed form submission
Extract text from web pages
Web page URL or loaded page → Scraped text content
Take screenshots of web pages
Web page URL or loaded page → Screenshot image file
Wait for page elements to load
Element selector and timeout duration → Element availability confirmation
Execute JavaScript on web pages
JavaScript code and target page → Script execution results
Monitor web page changes
Web page URL and monitoring criteria → Change detection notifications
Tags
Tech Stack
Media
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.