Connecting Mergado with AI agents (ChatGPT, Claude, Cursor)#

What is MCP and why use it?#

MCP (Model Context Protocol) is a standard that lets AI agents (ChatGPT, Claude, Cursor and others) work with your data and services. Once connected via MCP, an AI agent can on your behalf read your projects, rules and statistics in Mergado and, based on your instructions, act on them – all from within the AI conversation.

Instead of manually copying data from Mergado into the AI and back, the AI picks it up itself and returns the result directly.

Examples of what you can do with the AI once connected#

• “How many products does each of my projects currently have?”
• “I need to see all products that don’t have an EAN filled in.”
• “Run a feed audit on project XY and summarize what needs to be fixed.”
• “I need to change the product title so it follows the pattern: manufacturer + product name + size + color.”

What the Mergado MCP server can do#

Once connected, the AI agent gets access to the following data and operations in Mergado:

What the AI is allowed to do is determined by the permissions (scopes) of your access token or OAuth authorization – see Security. For example, if you grant the token only the “Read rules” scope, the AI will be able to read rules but not edit or delete them.

For the full list of individual permissions (scopes), see the Personal access token (PAT) article.

Connection options – overview#

There are two paths:

TL;DR: If you use ChatGPT or Claude via the official app and want a quick connection, wait for Path B (tentative release 27 May 2026). If you use Cursor, Claude Code, or are building your own integration, go straight with Path A.

Path A: Connect via PAT#

Prerequisites#

1. A generated PAT – see Personal access token (PAT).
2. Selected permissions (scopes) matching what you want the AI to do. For example, if you only want to read statistics, give the token only read scopes.
3. An MCP client (Cursor, Claude Code, Claude Desktop, custom script).

Common values for all clients#

Cursor#

1. Open Settings → MCP.
2. Click Add new MCP server.
3. Fill in:
   • Name: Mergado
   • URL: https://mcp.mergado.com
   • Header: Authorization: Bearer mergado_pat_…
4. Save and restart Cursor.
5. In the chat, check that the Mergado tools appear in the list of available tools.

Claude Code (CLI)#

claude mcp add --transport http mergado https://mcp.mergado.com \
  --header "Authorization: Bearer mergado_pat_..."

Check that the server is connected:

claude mcp list

Claude Desktop#

Claude Desktop currently does not support connecting to remote MCP servers directly over HTTP – it only communicates via local processes. You therefore need to use the mcp-remote helper, which acts as a local “bridge” between Claude Desktop and the remote mcp.mergado.com server.

1. Install Node.js#

mcp-remote runs on Node.js, which you need to have installed on your computer.

What is Node.js? A runtime environment that runs JavaScript programs on your computer. Some tools like mcp-remote are distributed as Node.js packages and cannot be started without it.

Download the installer from nodejs.org/en/download/current and choose the right variant based on:

• operating system (Windows / macOS / Linux),
• processor architecture – on macOS select ARM64 for Apple Silicon (M1 or later) processors, or x64 for older Intel-based Macs.

After installation, verify in a terminal (Command Prompt / Terminal):

node --version

If it prints a version (e.g. v22.x.x), installation succeeded.

2. Edit the Claude Desktop config file#

Open (or create) the file:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

And paste in the following configuration:

{
  "mcpServers": {
    "mergado": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://mcp.mergado.com",
        "--header",
        "Authorization: Bearer mergado_pat_..."
      ]
    }
  }
}

Replace mergado_pat_... with your full PAT.

3. Restart Claude Desktop#

After restarting, the Mergado MCP server should appear in Claude Desktop as an available tool. The first launch may take a few extra seconds – npx downloads the mcp-remote package on first run.

Tip for paid Claude plans (Pro, Max, Team, Enterprise): Claude Desktop also offers Custom Connectors (Settings → Connectors → Add custom connector), which supports remote MCP servers natively without Node.js and mcp-remote. If you have a paid plan, this is the easier route.

Custom script / application#

If you’re building your own MCP client or using an SDK (TypeScript, Python), connect to https://mcp.mergado.com with the Authorization: Bearer <PAT> header. Details of the MCP protocol are in the official MCP specification.

Path B: Official Mergado MCP App#

🚧 Currently in development. Tentative release: 27 May 2026. You’ll find updates in Mergado, on forum.mergado.com, or in this article, which we’ll update with the detailed steps as soon as the app is available.

After release, Mergado will be available as an official app directly in the connector directories of ChatGPT and Claude. Connecting will take two clicks – find Mergado in the connector list in your AI of choice, click Connect, and confirm the Mergado sign-in screen.

Compared to Path A, you won’t need to:

• generate or manage a personal access token (PAT),
• install Node.js or any other helper tools,
• manually edit configuration files.

Authorization works via the standard OAuth flow – just like when you sign in to a web service “with Google” or “with Facebook”. The Mergado MCP App will ask you to consent to specific permissions, and you can revoke them at any time in API and Access.

ChatGPT#

We’ll add the step-by-step instructions for ChatGPT once the official Mergado MCP App is released.

Claude (Connectors Directory)#

We’ll add the step-by-step instructions for Claude once the official Mergado MCP App is released.

What to do until then?#

Use Path A – for Claude it works right away through Claude Desktop or Claude Code; for Cursor it works through the built-in MCP integration.

Security#

An AI agent connected to Mergado acts on your behalf. When you say “delete the old rules”, it actually deletes the rules. A few principles to keep this under control:

Principle of least privilege#

Grant the token / OAuth authorization only the scopes the AI actually needs. If you only want to analyze statistics, don’t grant write access. That way the AI wouldn’t be able to change anything in Mergado even if it wanted to or received injected instructions.

Path A vs Path B from a security standpoint#

• Path A (PAT): the token is physically stored in the client’s configuration. If the configuration file falls into the wrong hands (Git, a shared screenshot, a stolen laptop), you have to revoke the token.
• Path B (OAuth): access is tied to a specific service (ChatGPT/Claude account) and you don’t use a “copyable” token. From a security standpoint this is more convenient.

Regularly review your access#

• For Path A, open API and Access in Mergado and revoke tokens you no longer use.
• For Path B, the same page will show the list of connected applications and let you disconnect them.

Watch out for prompt injection#

If you let an AI agent process external content (emails, web pages, product descriptions), it may contain a hidden instruction trying to manipulate the AI. Therefore:

• Grant write access sparingly and only where you really need it.
• For sensitive actions (deletion, bulk edits) ask the AI to confirm intent before executing.
• After the AI is done, verify the result in Mergado – what the AI did is visible in your project/rules as usual.

FAQ#

My token doesn’t work in Cursor / Claude Code, I’m getting 401. What to do?#

• Verify the header is exactly Authorization: Bearer mergado_pat_... (the word Bearer followed by a space).
• The token in Mergado must not be revoked and must not have expired – see the token list at API and Access.
• In Cursor/Claude Code, restart the application after changing the configuration.

The AI doesn’t have access to what I need. What to do?#

The token doesn’t have the right scope. Create a new token with broader permissions (an existing token cannot be expanded) and update it in the MCP client.

Can I use one PAT across multiple AI tools at the same time?#

Technically yes, but we recommend creating a separate token for each tool. When you want to revoke access from one tool, you only revoke that one token; the others keep working.

Path B isn’t available yet – can I use ChatGPT with Mergado in the meantime?#

Not via native integration in ChatGPT – fully click-free connection there will only be available after the official Mergado MCP App is released. In the meantime we recommend either waiting, or using a different AI agent where Path A works today: Claude Desktop, Claude Code or Cursor.

Will the Mergado MCP App be free?#

Yes. The official Mergado MCP App will be available for free to all Mergado users. Running your own MCP clients (Path A) is also free – you only pay for the AI tool you’re using (ChatGPT Plus, Claude Pro, etc.).

Which AI models are supported?#

Any MCP-compatible client. Today that primarily means: ChatGPT (Apps), Claude (Desktop, Code, Connectors), Cursor, and custom implementations using the MCP SDK.

Related links#

• Personal access token (PAT) – what it is and how to use it
• API and Access – manage PATs and connected applications
• Mergado API reference: api-docs.mergado.com
• MCP specification: modelcontextprotocol.io