AdButler MCP Server

Manage your entire AdButler account from any AI assistant — Claude, ChatGPT, Cursor, Windsurf, Cline, and any other Model Context Protocol client.

This MCP server exposes the full AdButler v2 API — 600+ tools covering advertisers, campaigns, zones, creatives, placements, VAST video ads, programmatic / RTB, reporting, targeting, drafts, contracts, product catalogs, and more — plus 9 pre-built workflow prompts that walk an AI through common tasks like launching a campaign or setting up retail media.

What you can ask

"Create a new campaign for Pepsi targeting users in Canada with a $5,000 lifetime budget, and assign it to my Homepage Banner zone."

"Show me the top 10 underperforming ad items in the last 7 days by CTR."

"Set up a VAST 4.2 pre-roll campaign with a 30-second skippable creative and three companion banners."

"Walk me through creating a new programmatic deal."

"Audit my ad units — which zones have no active placements?"

The AI translates these into the right sequence of AdButler API calls, runs them, and shows you the result.

Install

You have two options. Most users want the hosted version — zero setup, just paste a URL.

Option A — Hosted (recommended)

Use AdButler's hosted MCP server. No install, no Node, no npm.

The hosted server speaks both modern Streamable HTTP (/mcp) and legacy SSE (/sse) transports. New clients should prefer Streamable HTTP; SSE remains for backward compatibility.

Option B — Local stdio (npm)

Run the server locally as a Node process. Useful if you want to keep your API key out of any external service or run against a self-hosted AdButler.

Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json (macOS) or %APPDATA%\Claude\claude_desktop_config.json (Windows):

{
  "mcpServers": {
    "adbutler": {
      "command": "npx",
      "args": ["-y", "@adbutler/mcp-server"],
      "env": {
        "ADBUTLER_API_KEY": "your_api_key_here"
      }
    }
  }
}

Claude Code

claude mcp add adbutler --env ADBUTLER_API_KEY=your_api_key_here -- npx -y @adbutler/mcp-server

Cursor / Windsurf / Cline

Add to the client's MCP config:

{
  "mcpServers": {
    "adbutler": {
      "command": "npx",
      "args": ["-y", "@adbutler/mcp-server"],
      "env": { "ADBUTLER_API_KEY": "your_api_key_here" }
    }
  }
}

Get your AdButler API key

AdButler Dashboard → Settings → API Keys → create a new key. It's the same key the AdButler v2 REST API uses.

Authentication

The AdButler MCP works for both new and existing customers. What you see depends on whether you've already configured an API key.

Already an AdButler customer? (existing API key)

Use one of these:

Option 1 — set the API key in your connection config (recommended: persists across sessions)

Pass the key at connect time and every session authenticates automatically — the key never has to be shared in the chat.

• Claude Code — claude mcp add adbutler --transport http https://mcp.adbutler.com/mcp --header "Authorization: Bearer YOUR_API_KEY"
• Claude Desktop — edit claude_desktop_config.json → add headers: { "Authorization": "Bearer YOUR_API_KEY" } next to the server URL
• Cursor — MCP server settings → Headers → add Authorization: Bearer YOUR_API_KEY
• claude.ai custom connectors (no custom-header support) — use the URL form: https://mcp.adbutler.com/mcp?api_key=YOUR_API_KEY
• Any client — Authorization: Bearer … header or ?api_key=… query param, on /mcp (Streamable HTTP) or /sse (legacy SSE)

Reconnect — all 600+ tools become available immediately.

Option 2 — configure inside the chat (current session only on hosted)

Connect to the hosted MCP without auth and you'll see four onboarding tools. Then in the chat say:

"Configure my AdButler API key: your_api_key_here"

Your AI will call the setup_api_key tool, validate the key against your account, and unlock all the AdButler tools. You may need to disconnect and reconnect to refresh your client's tool list after.

⚠️ On the hosted server, a key configured this way lasts only until the session ends — a new chat, a client restart, or a server-side reconnect will ask for it again. If you find yourself pasting the key into the chat repeatedly (and your AI warning you about sharing secrets in conversation), switch to Option 1: configure the key once in the connection settings and it persists.

Troubleshooting: "my API key won't stay configured"

This means the key is being entered in-chat (Option 2) against the hosted server, where it is deliberately session-scoped — the hosted server is multi-tenant and never writes your key to disk. The fix is one-time setup in your client's connection config (Option 1 above): header for Claude Code / Claude Desktop / Cursor, ?api_key= URL parameter for claude.ai custom connectors. After that, reconnects and new chats authenticate automatically.

New to AdButler? (no account yet)

Connect to the hosted MCP without auth. The chat will guide you through a free trial signup using create_trial_account → check your email → verify_trial_email with the code. The API key is configured automatically.

Local stdio install

If you're running locally via npx, set ADBUTLER_API_KEY in the env block of your client's MCP config (see Option B above). The setup tools also work — and on local the key is remembered between sessions.

What's included

9 workflow prompts

Pre-built skill prompts that guide the AI through complete workflows end-to-end. Invoke them as MCP prompts (/launch-campaign etc.) from your client.

3 fallback meta-tools (search, describe, call)

When you need an endpoint that no specific tool wraps — or when the user asks about a field (e.g. "priority", "frequency cap") rather than a resource — three escape-hatch tools cover the gap:

These complement (don't replace) the specific tools below — the LLM continues to prefer specific tools when they obviously match.

600+ tools across the full AdButler API

100% coverage of the AdButler v2 OpenAPI spec (604/604 endpoints).

Telemetry

The hosted server at mcp.adbutler.com collects usage analytics so we can understand which tools are popular, prioritize improvements, and diagnose failures. We capture, per tool call: