Mcp E2b
Groq + E2B MCP Gateway: Multiple MCP Servers in One Sandbox
This notebook demonstrates how to use E2B's MCP Gateway to run multiple Model Context Protocol servers together with Groq's ultra-fast inference.
What is E2B MCP Gateway?
E2B's MCP Gateway provides a unified interface to multiple MCP servers running inside an isolated sandbox. Instead of connecting to each MCP server separately, you get:
- One endpoint for all your MCP tools (Exa, Browserbase, Notion, etc.)
- Isolated environment where MCPs run securely
- Seamless integration with Groq's Responses API
- 200+ tools from the Docker MCP Catalog
What We'll Build
We'll create an AI research agent that:
- Searches for AI companies using Exa MCP's web search
- Visits their websites using Browserbase MCP's browser automation
- Extracts pricing and contact information
- Returns structured results—all through one E2B gateway
This showcases how E2B makes it trivial to combine multiple MCP capabilities in a single workflow.
Getting Started
You'll need API keys from four services:
- E2B (e2b.dev) - Provides the unified MCP gateway and sandbox
- Browserbase (browserbase.com) - Browser automation MCP
- You'll need both an API key AND a Project ID
- Exa (dashboard.exa.ai) - Web search MCP
- Groq (console.groq.com) - Ultra-fast LLM inference
Set Environment Variables
After obtaining your API keys, set them as environment variables:
macOS/Linux:
export E2B_API_KEY='your_e2b_key'
export BROWSERBASE_API_KEY='your_browserbase_key'
export BROWSERBASE_PROJECT_ID='your_project_id'
export EXA_API_KEY='your_exa_key'
export GROQ_API_KEY='your_groq_key'
Windows (PowerShell):
$Env:E2B_API_KEY='your_e2b_key'
$Env:BROWSERBASE_API_KEY='your_browserbase_key'
$Env:BROWSERBASE_PROJECT_ID='your_project_id'
$Env:EXA_API_KEY='your_exa_key'
$Env:GROQ_API_KEY='your_groq_key'
Installation
Install the required packages:
Setup: Import Libraries and Configure API Keys
Step 1: Create E2B Sandbox with MCP Gateway
Here's where the magic happens. We create a single E2B sandbox that hosts both Exa and Browserbase MCP servers:
Step 2: Setup Groq Client
Initialize the Groq client using OpenAI's SDK (Groq is OpenAI-compatible):
Step 3: Configure MCP Tools for Groq
Tell Groq about the E2B MCP gateway. Notice we only need ONE tool configuration—E2B routes to all MCPs internally:
Core Function: Research with Multiple MCPs
This function demonstrates the power of E2B's unified gateway:
Demo 1: Research AI Inference Companies
Let's research AI companies that provide inference APIs and extract their pricing information.
Watch how the agent:
- Uses Exa to search for AI inference companies
- Uses Browserbase to visit their websites
- Extracts pricing and contact details
All of this happens through the single E2B gateway we configured:
Display the response in markdown format:
Demo 2: Custom Research Task
Now it's your turn! Replace the query below with your own research task.
Ideas to try:
- "Find popular AI image generation tools and check their pricing"
- "Research vector database companies and extract their key features"
- "Find AI agent frameworks and get their documentation links"
- "Research LLM observability tools and compare their features"
Next Steps
Now that you've seen E2B's MCP gateway in action, here are some ideas to explore:
Add More MCP Servers
E2B supports 200+ MCP servers from the Docker MCP Catalog. Try adding:
- Notion - Store research findings in a database
- Airtable - Organize data in spreadsheets
- GitHub - Search code repositories
- Slack - Send notifications
- Firecrawl - Advanced web scraping
Just add them to the mcp configuration:
sandbox = Sandbox.create(
mcp={
'browserbase': {...},
'exa': {...},
'notion': {
'internalIntegrationToken': NOTION_API_KEY,
},
# Add more...
}
)
Build Production Workflows
- Create scheduled research jobs that run daily
- Build a research API endpoint
- Set up monitoring and alerting
- Store results in a database
Scale with Multiple Sandboxes
- Run parallel research tasks in separate sandboxes
- Implement rate limiting and queuing
- Add retry logic and error handling
Experiment with Different Models
Try different Groq models for different tasks:
llama-3.3-70b-versatile- Great for general tasksqwen/qwen3-32b- Fast and efficientopenai/gpt-oss-120b- Powerful reasoning
Optimize for Speed
- Use streaming responses for real-time updates
- Batch similar requests together
- Cache frequently accessed data
Resources
E2B MCP Gateway
- Documentation: e2b.dev/docs/mcp
- Quickstart: e2b.dev/docs/mcp/quickstart
- Available Servers: e2b.dev/docs/mcp/available-servers
- Custom Servers: e2b.dev/docs/mcp/custom-servers
- GitHub: github.com/e2b-dev/e2b
MCP Servers Used
- Exa MCP: docs.exa.ai/reference/exa-mcp
- Exa GitHub: github.com/exa-labs/exa-mcp-server
- Browserbase MCP: smithery.ai/server/@browserbasehq/mcp-browserbase
- Stagehand Docs: docs.stagehand.dev
Groq
- API Documentation: console.groq.com/docs
- Models: console.groq.com/docs/models
- Cookbook: github.com/groq/groq-api-cookbook
Model Context Protocol
- Official Site: modelcontextprotocol.io
- Specification: spec.modelcontextprotocol.io
- GitHub: github.com/modelcontextprotocol
Happy researching with E2B + Groq! 🚀