​
MCP Server Implementation

This guide provides detailed instructions for integrating Co-Browser’s MCP server into your AI agents, enabling them to safely access web content and APIs through our authentication system.

​
Implementation Options

Co-Browser’s MCP server can be integrated into your AI systems in two main ways:

The simplest way to implement MCP is directly through the Cursor client, providing your AI agent with expanded capabilities.

​
Python SDK Integration

​
Installation

Install the MCP client package using pip:

pip install mcp-client

​
Basic Setup

Here’s the basic setup for connecting your AI agent to our MCP server using the Python SDK:

import asyncio
import logging

from mcp import ClientSession
from mcp.client.sse import sse_client

logging.basicConfig(
    level=logging.INFO, format="%(asctime)s - %(levelname)s - %(message)s"
)

async def main():
    async with sse_client(
        url="https://api-gateway.cobrowser.xyz/mcp/api/sse?api_key=YOUR_API_KEY",
    ) as streams:
        async with ClientSession(*streams) as session:
            await session.initialize()

            # List available tools for your AI agent
            tools = await session.list_tools()
            print(tools)

            # Add your AI agent's MCP operations here

asyncio.run(main())

Replace YOUR_API_KEY with the API key you generated in the Developer Dashboard.

​
Browser-Use Example

This example demonstrates how to enable your AI agent to navigate to a website and extract content:

import asyncio
import logging

from mcp import ClientSession
from mcp.client.sse import sse_client

logging.basicConfig(
    level=logging.INFO, format="%(asctime)s - %(levelname)s - %(message)s"
)

async def main():
    async with sse_client(
        url="https://api-gateway.cobrowser.xyz/mcp/browser/sse?api_key=YOUR_API_KEY",
    ) as streams:
        async with ClientSession(*streams) as session:
            await session.initialize()

            # Navigate to a website
            response = await session.invoke(
                name="browser_use",
                params={
                    "action": "navigate",
                    "url": "https://example.com"
                }
            )
            print(f"Navigation response: {response}")

            # Extract content from the page
            content = await session.invoke(
                name="browser_use",
                params={
                    "action": "getContent",
                    "selector": "h1"
                }
            )
            print(f"Page heading: {content}")

asyncio.run(main())

​
Gmail Operations Example

This example shows how to enable your AI agent to read and send emails:

import asyncio
import logging

from mcp import ClientSession
from mcp.client.sse import sse_client

logging.basicConfig(
    level=logging.INFO, format="%(asctime)s - %(levelname)s - %(message)s"
)

async def main():
    async with sse_client(
        url="https://api-gateway.cobrowser.xyz/mcp/api/sse?api_key=YOUR_API_KEY",
    ) as streams:
        async with ClientSession(*streams) as session:
            await session.initialize()

            # AI agent reads emails with a specific subject
            emails = await session.invoke(
                name="gmail_read",
                params={
                    "subject": "Important Update",
                    "maxResults": 3
                }
            )
            print(f"Found {len(emails)} emails")

            # AI agent processes email content
            for email in emails:
                print(f"Subject: {email['subject']}")
                print(f"Snippet: {email['snippet']}")
                print("---")

            # AI agent sends an email based on processed information
            result = await session.invoke(
                name="gmail_write",
                params={
                    "destination": "[email protected]",
                    "subject": "Response from AI Assistant",
                    "body": "This is a response generated by your AI assistant based on analyzed emails."
                }
            )
            print(f"Email sent: {result}")

asyncio.run(main())

​
Error Handling

Implement robust error handling to ensure your AI agent gracefully manages issues:

import asyncio
import logging
from mcp import ClientSession, MCPError
from mcp.client.sse import sse_client

logging.basicConfig(
    level=logging.INFO, format="%(asctime)s - %(levelname)s - %(message)s"
)

async def main():
    try:
        async with sse_client(
            url="https://api-gateway.cobrowser.xyz/mcp/api/sse?api_key=YOUR_API_KEY",
        ) as streams:
            async with ClientSession(*streams) as session:
                await session.initialize()

                try:
                    # AI agent attempts an operation
                    result = await session.invoke(
                        name="gmail_read",
                        params={
                            "subject": "Important Update"
                        }
                    )
                    print(result)

                except MCPError as e:
                    # Handle operation-specific errors
                    print(f"Operation error: {e}")
                    # Implement fallback behavior for your AI agent

    except Exception as e:
        # Handle connection errors
        print(f"Connection error: {e}")
        # Implement fallback behavior for your AI agent

asyncio.run(main())

​
Permission Controls

For responsible AI agent implementation, configure appropriate permission settings:

import asyncio
import logging
from mcp import ClientSession, ClientConfig, PermissionSettings
from mcp.client.sse import sse_client

logging.basicConfig(
    level=logging.INFO, format="%(asctime)s - %(levelname)s - %(message)s"
)

async def main():
    # Define permission settings for your AI agent
    permissions = PermissionSettings(
        allow_browser_navigation=True,
        allow_email_reading=True,
        allow_email_sending=False,  # Restrict email sending
        require_user_confirmation=True  # Require user confirmation for critical actions
    )

    # Custom configuration with permissions
    config = ClientConfig(
        timeout=30,
        retry_attempts=3,
        permissions=permissions
    )

    async with sse_client(
        url="https://api-gateway.cobrowser.xyz/mcp/api/sse?api_key=YOUR_API_KEY",
    ) as streams:
        async with ClientSession(*streams, config=config) as session:
            await session.initialize()

            # Your AI agent's operations here

asyncio.run(main())

​
AI Agent Integration Best Practices

  1. Implement confirmation flows for critical actions
  2. Provide transparency about what your AI agent is accessing
  3. Handle failures gracefully with appropriate fallback behaviors
  4. Log operations for auditing and improvement
  5. Limit permission scopes to only what your agent needs

​
Next Steps

  • Explore the MCP Services documentation for all available capabilities
  • Visit the Developer Dashboard to manage API keys and permissions
  • Contact support if you have any implementation questions