Skip to content
AgentBox Documentation

MCP Gateway

What is MCP Gateway?

Section titled “What is MCP Gateway?”

MCP Gateway is a service that enables you to run and manage multiple Model Context Protocol (MCP) servers within your AgentBox sandbox. It provides a unified gateway interface to access various third-party services and tools through a single endpoint.

Supported MCP Servers

Section titled “Supported MCP Servers”

MCP Gateway supports many different MCP servers, including:

  • Search & Web: DuckDuckGo, Brave, Exa, ArXiv, Wikipedia
  • Databases: PostgreSQL, MySQL, MongoDB, Redis, Chroma, Pinecone
  • Cloud Services: AWS, Azure, GCP, Vercel, Heroku
  • Development Tools: GitHub, GitLab, Code Interpreter, Filesystem
  • AI Services: OpenAI, Anthropic, Hugging Face, ElevenLabs
  • Communication: Slack, Discord, Gmail, Notion
  • And many more…

Browse all available servers →

Create a sandbox with MCP Gateway

Section titled “Create a sandbox with MCP Gateway”

To create a sandbox with MCP Gateway, specify the MCP configuration when creating the sandbox. The mcp-gateway template will be automatically selected.

Basic usage

Section titled “Basic usage”

Create a sandbox with a single MCP server:

from agentbox import Sandbox

sandbox = Sandbox.create(
   api_key="ab_xxxxxxxxxxxxxxxxxxxxxxxx",
   mcp={
       "duckduckgo": {}
   }
)

Multiple MCP servers

Section titled “Multiple MCP servers”

You can configure multiple MCP servers in a single sandbox:

from agentbox import Sandbox

sandbox = Sandbox.create(
   api_key="ab_xxxxxxxxxxxxxxxxxxxxxxxx",
   mcp={
       "duckduckgo": {},
       "arxiv": {"storagePath": "/tmp/arxiv"},
       "context7": {}
   }
)

Get MCP Gateway URL and Token

Section titled “Get MCP Gateway URL and Token”

After creating a sandbox with MCP Gateway, you can retrieve the URL and token to connect to it:

from agentbox import Sandbox

sandbox = Sandbox(api_key="ab_xxxxxxxxxxxxxxxxxxxxxxxx",
                 sandbox_id="your-sandbox-id")

# Get MCP connection info
mcp_url = sandbox.get_mcp_url()
mcp_token = sandbox.get_mcp_token()

print(f"MCP URL: {mcp_url}")
print(f"MCP Token: {mcp_token}")

The URL format is: https://{sandbox-domain}:50005/mcp

Connect to MCP Gateway

Section titled “Connect to MCP Gateway”

Once you have the MCP URL and token, you can connect to it from various MCP clients.

Connect with Python

Section titled “Connect with Python”

Using the official MCP Python client:

from mcp.client.session import ClientSession
from mcp.client.streamable_http import streamablehttp_client
from datetime import timedelta

async def connect_to_mcp(mcp_url, mcp_token):
   async with streamablehttp_client(
       url=mcp_url,
       headers={"Authorization": f"Bearer {mcp_token}"},
       timeout=timedelta(seconds=600)
   ) as (read_stream, write_stream, _):
       async with ClientSession(read_stream, write_stream) as session:
           await session.initialize()
           
           # List available tools
           tools = await session.list_tools()
           print(f"Available tools: {[tool.name for tool in tools.tools]}")
           
           # Use a tool
           result = await session.call_tool("duckduckgo_search", {
               "query": "AgentBox MCP Gateway"
           })
           print(result)

Connect with TypeScript/JavaScript

Section titled “Connect with TypeScript/JavaScript”

Using the official MCP TypeScript client:

import { Client } from "@modelcontextprotocol/sdk/client/index.js";
import { SSEClientTransport } from "@modelcontextprotocol/sdk/client/sse.js";

async function connectToMCP(mcpUrl: string, mcpToken: string) {
 const transport = new SSEClientTransport(
   new URL(mcpUrl),
   {
     headers: {
       Authorization: `Bearer ${mcpToken}`,
     },
   }
 );

 const client = new Client(
   {
     name: "agentbox-mcp-client",
     version: "1.0.0",
   },
   {
     capabilities: {},
   }
 );

 await client.connect(transport);

 // List available tools
 const tools = await client.listTools();
 console.log("Available tools:", tools.tools.map(t => t.name));

 // Use a tool
 const result = await client.callTool({
   name: "duckduckgo_search",
   arguments: { query: "AgentBox MCP Gateway" },
 });
 console.log(result);
}

Configure Claude Desktop

Section titled “Configure Claude Desktop”

To use MCP Gateway with Claude Desktop, add it to your Claude Desktop configuration:

// ~/Library/Application Support/Claude/claude_desktop_config.json
{
 "mcpServers": {
   "agentbox-mcp-gateway": {
     "transport": {
       "type": "sse",
       "url": "https://sandbox-abc123.agentbox.cloud:50005/mcp",
       "headers": {
         "Authorization": "Bearer your-mcp-token"
       }
     }
   }
 }
}

Troubleshooting

Section titled “Troubleshooting”

Connection timeout: Make sure your sandbox is running and not paused.

Authentication error: Verify that you’re using the correct token and that it hasn’t expired.

Server not found: Ensure the MCP server is properly configured in your sandbox.

Next Steps

Section titled “Next Steps”