MCP Configure

Configure Attribuly's MCP server to get AI assistance with your Attribuly data.

You will learn

The Attribuly Model Context Protocol (MCP) server seamlessly integrates with Attribuly’s APIs, enabling AI clients to interact with your Attribuly data. This guide will walk you through configuring various MCP clients for both the remote-hosted MCP server (recommended) and the local MCP server.

MCP (Model Context Protocol) is an open protocol that helps AI models securely interface with different data sources and tools. You’ll connect Attribuly’s MCP server to an MCP client, i.e., an AI agent like Claude or Cursor, to get assistance with your Attribuly data. Your MCP client can help with a variety of tasks from campaign creation to performance reports. Here are some examples of prompts you can give:

Show me the performance of my campaigns from the last 30 days.

Which campaign are performing the best in terms of conversions?

Configuration

MCP Server Address (https://data.api.attribuly.com/mcp)

For authentication, please use your API key. For details, refer to the API Authentication Guide.

Claude Code CLI

Install Claude Code if you haven't already, then verify in your terminal:

Add Attribuly as a remote MCP server. Use --scope user to make it available across all projects (drop the flag if you only want it inside the current directory):

Verify it was added:

You should see attribuly: https://data.api.attribuly.com/mcp (HTTP) - ✓ Connected.

Start a Claude Code session and try a prompt that uses your data, e.g. "Use Attribuly to show the top 5 campaigns by ROAS for the last 30 days."

Note: A ✓ Connected status only confirms the endpoint is reachable. Your API key is validated when a tool is actually invoked, so the first real query will surface any auth errors.

Claude Desktop

Download Claude Desktop.

Make sure Node.js is installed (required for mcp-remote). Verify in your terminal:

node -v

Open the Claude Desktop config file (create it if it doesn't exist):

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json

Windows: %APPDATA%\Claude\claude_desktop_config.json

Add the following, substituting your API key. Keep any existing preferences block intact.

macOS / Linux:

{
  "mcpServers": {
    "attribuly": {
      "command": "npx",
      "args": [
        "mcp-remote",
        "https://data.api.attribuly.com/mcp",
        "--header",
        "ApiKey: your-api-key"
      ]
    }
  }
}

Windows — wrap the command with cmd /c and add -y. Using bare npx on Windows commonly causes a "Server disconnected" error (see Troubleshooting below):

{
  "mcpServers": {
    "attribuly": {
      "command": "cmd",
      "args": [
        "/c",
        "npx",
        "-y",
        "mcp-remote",
        "https://data.api.attribuly.com/mcp",
        "--header",
        "ApiKey: your-api-key"
      ]
    }
  }
}

Save the file, fully quit Claude Desktop (menu-bar / system-tray icon → Quit, not just closing the window), then relaunch.

Verify: open Connectors in Claude Desktop — you should see attribuly under the Desktop group. You can also test the connection from your terminal:

npx mcp-remote https://data.api.attribuly.com/mcp --header "ApiKey: your-api-key"

A successful run prints:

Proxy established successfully between local STDIO and remote StreamableHTTPClientTransport

Troubleshooting

"Server disconnected" on Windows.
This is almost always caused by how Windows launches npx, not by your API key or network. Two things cause it:

On Windows, npx is npx.cmd and isn't spawned through a shell by default, so the process can exit immediately.

Without -y, the first run prompts Need to install the following packages: mcp-remote. Ok to proceed? (y). Claude Desktop starts the server silently and can't answer the prompt, so it hangs and disconnects.

Fix: use the cmd /c npx -y … form shown in step 3.

To confirm the underlying connection is fine, run the terminal test command from step 5. If it prints Proxy established successfully, your API key and network are working — the disconnect is purely the launch-command issue.

Optional (most robust): install mcp-remote globally with npm i -g mcp-remote, then set "command": "mcp-remote" and drop npx/cmd from args entirely.

"Logged out" errors on claude.ai web.
The Claude.ai web Custom Connector UI does not support custom request headers, so it cannot pass the required ApiKey header. Use Claude Desktop (not claude.ai web) for this configuration.

Old keyless connection takes precedence.
If you previously added Attribuly on claude.ai web: go to claude.ai → Settings → Connectors, find Attribuly Custom, click ⋮ → Disconnect. Otherwise Claude will prefer the old keyless connection and fail.

Cursor

Download Cursor

Open Cursor. Within Settings > Cursor Settings > Tools & Integrations > Add Custom MCP, add the following, substituting your API key and MCP server address:

{
  "mcpServers": {
    "attribuly": {
      "type": "http",
      "url": "https://data.api.attribuly.com/mcp",
      "headers": {
        "ApiKey": "your-api-key"
      }
    }
  }
}

Once you have saved all configuration changes, terminate all Cursor processes and relaunch Cursor to apply the new settings.

Verify the connection by looking for visual indicators and asking a question to test the connection.

VS Code

Download VS Code

Open VS Code. Within settings (Ctrl + Shift + P > Preferences: Open Settings (JSON)), add the following, substituting your API key and MCP server address:

{
  "mcp": {
    "servers": {
      "attribuly": {
        "type": "http",
        "url": "https://data.api.attribuly.com/mcp",
        "headers": {
          "ApiKey": "your-api-key"
        }
      }
    }
  }
}

Once you have saved all configuration changes, terminate all VS Code processes and relaunch VS Code to apply the new settings.

Verify the connection by looking for visual indicators and asking a question to test the connection.

ChatGPT

1. Enable Developer Mode

Open ChatGPT (web/desktop app).

Navigate to Settings → Apps & Connectors.

Under the Advanced section, toggle Developer Mode to enabled.

https://developers.openai.com/apps-sdk/deploy/connect-chatgpt

2. Create Connector

In Settings → Connectors, click the Create button.

Fill in the configuration form:

Name: Attribuly

Server URL: https://data.api.attribuly.com/mcp

Check the box: I trust this provider

Add authentication (ApiKey)

Click Create to finish connector setup.

Note: Without Developer Mode enabled, ChatGPT will reject the server if it lacks search/fetch tools. With Developer Mode on, search/fetch tools are not required for Chat mode usage.

3. Use in Chat

Start a new chat conversation in ChatGPT.

Click the + button → select More → choose Developer Mode.

Enable your MCP server connector (critical: the connector must be explicitly added to each chat session).

Now you can use the tools provided by your MCP server in the chat.

Available Tools

Tool Name	Description
get_all_attribution_data	Get omnichannel attribution report data
get_ad_attribution_data	Get ad attribution report data
get_product_analysis_data	Get product analysis report data
get_influencer_data	Get influencer attribution report data
get_link_builder_data	Get custom tracking link attribution report data

Getting Started

Create an account

Get your private API key from Settings > API Keys

Configure your MCP client following the instructions above

Start using your MCP client

Configure Attribuly's MCP server to get AI assistance with your Attribuly data.#

You will learn#

Configuration#

Claude Code CLI#

Claude Desktop#

Troubleshooting#

Cursor#

VS Code#

ChatGPT#

1. Enable Developer Mode#

2. Create Connector#

3. Use in Chat#

Available Tools#

Getting Started#