Mcp Server

What is Mcp Server about?

Mcp Server powers Integration.app by exposing integration‑specific tools (actions) via MCP endpoints. It can operate in static mode—returning every available tool for the authenticated user—or in dynamic mode, where a single enable-tools tool is offered to selectively activate needed actions.

How to use Mcp Server?

1. Install

   git clone https://github.com/integration-app/mcp-server.git
   cd mcp-server
   npm install
   npm run build
   

2. Run locally

   npm run dev
   

   The server listens on http://localhost:3000.
3. Connect from a client Use the recommended Streamable HTTP transport:

   import { Client } from '@modelcontextprotocol/sdk/client/index.js';
   import { StreamableHTTPClientTransport } from '@modelcontextprotocol/sdk/client/streamableHttp.js';
   
   const client = new Client({ name: 'example', version: '1.0.0' });
   const transport = new StreamableHTTPClientTransport(
     new URL('https://<HOSTED_MCP_SERVER_URL>/mcp?mode=dynamic'),
     { requestInit: { headers: { Authorization: `Bearer ${ACCESS_TOKEN}` } } }
   );
   await client.connect(transport);
   

4. Enable specific tools (dynamic mode)

   await client.callTool({
     name: 'enable-tools',
     arguments: { tools: ['gmail-send-email', 'gmail-read-email'] }
   });
   

5. Optional chat session tracking – add an x-chat-id header to persist a conversation across calls.

Key Features

• Two transports: deprecated SSE (/sse) and recommended Streamable HTTP (/mcp).
• Static & Dynamic modes – return all tools or enable selective tools on demand.
• App filtering – fetch tools for specific integrations via apps query parameter.
• Experimental chat session management via x-chat-id header and /mcp/sessions endpoint.
• Docker support for containerized deployment.
• Compatibility configurations for Cursor and Claude Desktop.

Use Cases

• Integrating Gmail, Google Calendar, or other SaaS APIs as selectable tools for LLM agents.
• Building AI assistants that need to call specific actions only when relevant, reducing token usage.
• Providing a unified tool catalogue to multiple AI platforms (e.g., Cursor, Claude) via a single MCP endpoint.
• Managing persistent conversational context across multiple LLM calls using chat IDs.

FAQ

Q: Which transport should I use? A: Use Streamable HTTP (/mcp) – it is the current, bidirectional, and recommended transport.

Q: How do I limit the tools returned? A: Deploy the server in dynamic mode (?mode=dynamic) and invoke the enable-tools tool with the desired list.

Q: Can I run the server in the cloud? A: Yes. Build the Docker image (docker build -t integration-app-mcp-server .) and run it on any container service.

Q: How do I fetch tools for only Google Calendar? A: Add apps=google-calendar to the query string, e.g., /mcp?apps=google-calendar.

Q: How is chat session persistence achieved? A: Include an x-chat-id header with a unique identifier; the server maintains a session mapped to that ID.

Integration App MCP Server

The Integration App MCP Server is a Model Context Protocol (MCP) server, it provides actions for connected integrations on Integration.app membrane as tools.

Here's our official AI Agent Example that shows you how to use this MCP server in your application.

📋 Prerequisites

• Node.js (v18 or higher)
• An Integration.app account

⚙️ Installation

git clone https://github.com/integration-app/mcp-server.git
cd mcp-server
npm install
npm run build

🛠️ Local Development

To run the development server locally, start it with:

npm run dev

The server will be live at http://localhost:3000 ⚡️

🧪 Running tests

# Run the server in test mode
npm run start:test

# then run tests
npm test

🚀 Deployment

Deploy your own instance of this MCP server to any cloud hosting service of your choice.

🐳 Docker

The project includes a Dockerfile for easy containerized deployment.

docker build -t integration-app-mcp-server .
docker run -p 3000:3000 integration-app-mcp-server

🔗 Connecting to the MCP server

This MCP server support two transports:

Transport	Endpoint	Status
SSE (Server‑Sent Events)	/sse	🔴 Deprecated — deprecated as of November 5, 2024 in MCP spec
HTTP (Streamable HTTP)	/mcp	🟢 Recommended — replaces SSE and supports bidirectional streaming

🔐 Authentication

Provide an Integration.app access token via query or Authorization header:

?token=ACCESS_TOKEN
Authorization: Bearer ACCESS_TOKEN

SSE (Deprecated)

await client.connect(
  new SSEClientTransport(
    new URL(
      `https://<HOSTED_MCP_SERVER_URL>/sse`
    )
    {
      requestInit: {
        headers: {
          Authorization: `Bearer ${ACCESS_TOKEN}`,
        },
      },
    }
  )
);

Streamable HTTP (Recommended)

await client.connect(
  new StreamableHTTPClientTransport(
    new URL(`https://<HOSTED_MCP_SERVER_URL>/mcp`)
    {
      requestInit: {
        headers: {
          Authorization: `Bearer ${ACCESS_TOKEN}`,
        },
      },
    }
  )
);

⚡ Static vs Dynamic Mode

By default, the MCP server runs in static mode, which means it returns all available tools (actions) for all connected integrations.

With dynamic mode (?mode=dynamic), the server will only return one tool: enable-tools. You can use this tool to selectively enable the tools you actually need for that session.

In dynamic mode, your implementation should figure out which tools are most relevant to the user's query. Once you've identified them, prompt the LLM to call the enable-tools tool with the appropriate list.

Want to see how this works in practice? Check out our AI Agent Example.

import { Client } from '@modelcontextprotocol/sdk/client/index.js';
import { StreamableHTTPClientTransport } from '@modelcontextprotocol/sdk/client/streamableHttp.js';

const client = new Client({
  name: 'example-integration-app-mcp-client',
  version: '1.0.0',
});

const transport = new StreamableHTTPClientTransport(
  new URL(`https://<HOSTED_MCP_SERVER_URL>/mcp?mode=dynamic`),
  {
    requestInit: {
      headers: {
        Authorization: `Bearer ${ACCESS_TOKEN}`,
      },
    },
  }
);

await client.connect(transport);

await client.callTool({
  name: 'enable-tools',
  arguments: {
    tools: ['gmail-send-email', 'gmail-read-email'],
  },
});

🔧 Getting tools for a specific integrations

In static mode, the MCP server fetches tools from all active connections associated with the provided token.

You can choose to only fetch tools for a specific integration by passing the apps query parameter: /mcp?apps=google-calendar,google-docs

💬 Chat Session Management (Experimental)

The MCP server (streamable-http transport only) supports persistent chat sessions. Include an x-chat-id header in your requests to automatically track sessions for that specific chat. This is an experimental feature that we provide in addition to standard MCP sessions.

Starting a new chat session:

POST /mcp
Authorization: Bearer YOUR_ACCESS_TOKEN
x-chat-id: my-awesome-chat-123

Retrieving your chat sessions:

GET /mcp/sessions
Authorization: Bearer YOUR_ACCESS_TOKEN

Response:

{
  "my-awesome-chat-123": "session-uuid-1",
  "another-chat-456": "session-uuid-2"
}

This feature lets you use same session for a conversation. Check out our AI Agent Example to see how this works in practice.

Configuring other MCP clients

📝 Cursor

To use this server with Cursor, update the ~/.cursor/mcp.json file:

{
  "mcpServers": {
    "integration-app": {
      "url": "https://<HOSTED_MCP_SERVER_URL>/sse?token={ACCESS_TOKEN}"
    }
  }
}

Restart Cursor for the changes to take effect.

🤖 Claude Desktop

To use this server with Claude, update the config file (Settings > Developer > Edit Config):

{
  "mcpServers": {
    "integration-app": {
      "url": "https://<HOSTED_MCP_SERVER_URL>/sse?token={ACCESS_TOKEN}"
    }
  }
}

🔧 Troubleshooting

• Ensure your access token is valid and you're generating it according to these instructions
• Check the MCP server logs for any errors or issues during startup or connection attempts.
• Use the MCP Inspector for testing and debugging