Browserbase MCP

Browserbase MCP (opens in a new tab) gives AI models actual browser control through cloud infrastructure. Built on Stagehand, it lets you navigate websites, extract data, and interact with web pages using natural language commands.

The integration provides real browser sessions that AI can control, enabling tasks that go beyond simple web search APIs.

Available Tools

Multi-Session Tools

• multi_browserbase_stagehand_session_create: Create parallel browser sessions
• multi_browserbase_stagehand_session_list: Track active sessions
• multi_browserbase_stagehand_session_close: Clean up sessions
• multi_browserbase_stagehand_navigate_session: Navigate in specific session

Core Browser Actions

• browserbase_stagehand_navigate: Navigate to URLs
• browserbase_stagehand_act: Perform actions ("click the login button")
• browserbase_stagehand_extract: Extract text content
• browserbase_stagehand_observe: Find page elements
• browserbase_screenshot: Capture screenshots

Session Management

• browserbase_session_create: Create or reuse sessions
• browserbase_session_close: Close active sessions

Prerequisites

• Jan with MCP enabled
• Browserbase account (includes 60 minutes free usage)
• Model with strong tool calling support
• Node.js installed

Setup

Enable MCP

1. Go to Settings > MCP Servers
2. Toggle Allow All MCP Tool Permission ON

Get Browserbase Credentials

1. Sign up at browserbase.com (opens in a new tab)

   • Email verification required
   • Phone number authentication
   • Thorough security process

2. Access your dashboard and copy:

   • API Key
   • Project ID

Configure MCP Server

Click + in MCP Servers section:

NPM Package Configuration:

• Server Name: browserbase
• Command: npx
• Arguments: @browserbasehq/mcp-server-browserbase
• Environment Variables:
  • Key: BROWSERBASE_API_KEY, Value: your-api-key
  • Key: BROWSERBASE_PROJECT_ID, Value: your-project-id

Verify Setup

Check the tools bubble in chat to confirm Browserbase tools are available:

Real Usage Example

Live Information Query

This simple query demonstrates browser automation in action:

1. Tool Activation
   • Model creates browser session
   • Navigates to sports websites
   • Extracts current match data

2. Results Delivery
   • Real-time match information
   • Multiple sports covered
   • Current scores and timings

The AI successfully found:

• AFL matches with live scores
• NRL games in progress
• Upcoming Rugby Union fixtures

Common Issues

Tool Call Failures

Sometimes tool calls fail due to parsing issues:

Solutions:

• Try rephrasing your prompt
• Disable unnecessary tools
• Use simpler, more direct requests
• Switch to Claude 3.5+ Sonnet if using another model

Model Limitations

Most models struggle with multiple tools. If experiencing issues:

• Start with single-purpose requests
• Build complexity gradually
• Consider which tools are actually needed
• Expect some trial and error initially

Usage Limits

Free Tier:

• 60 minutes of browser time included
• Sessions auto-terminate after 5 minutes inactivity
• Can adjust timeout in Browserbase dashboard
• Usage visible in dashboard analytics

Session Management:

• Each browser session counts against time
• Close sessions when done to conserve minutes
• Multi-session operations consume time faster

Practical Use Cases

Real-Time Data Collection

Form Testing

Content Monitoring

Multi-Site Analysis

Automated Verification

Advanced Techniques

Session Reuse

Parallel Operations

Sequential Workflows

Optimization Tips

Prompt Engineering:

• Be specific about what to extract
• Name exact websites when possible
• Break complex tasks into steps
• Specify output format clearly

Tool Selection:

• Use multi-session only when needed
• Close sessions promptly
• Choose observe before act when possible
• Screenshot sparingly to save time

Error Recovery:

• Have fallback prompts ready
• Start simple, add complexity
• Watch for timeout warnings
• Monitor usage in dashboard

Troubleshooting

Connection Issues:

• Verify API key and Project ID
• Check Browserbase service status
• Ensure NPX can download packages
• Restart Jan after configuration

Browser Failures:

• Some sites block automation
• Try different navigation paths
• Check if site requires login
• Verify target site is accessible

Performance Problems:

• Reduce concurrent sessions
• Simplify extraction requests
• Check remaining time quota
• Consider upgrading plan

Model Struggles:

• Too many tools overwhelm most models
• Claude 3.5+ Sonnet most reliable
• Reduce available tools if needed
• Use focused, clear instructions

Browserbase vs Browser Use

Next Steps

Browserbase MCP provides genuine browser automation capabilities, not just web search. This enables complex workflows like form filling, multi-site monitoring, and data extraction that would be impossible with traditional APIs.

The cloud infrastructure handles browser complexity while Jan maintains conversational privacy. Just remember: with great browser power comes occasional parsing errors.