Toffu MCP Server

The Toffu MCP server lets you run marketing work — campaign performance, creative reports, memory search, budget changes — from Claude, Cursor, ChatGPT, or any other MCP client. The server URL is https://mcp.toffu.ai/mcp, and most clients connect with one-click OAuth: no API key, no config file.

MCP (Model Context Protocol) is the open standard AI clients use to talk to external tools. Toffu supports it in both directions: your AI client can call Toffu, and Toffu can call MCP servers you add.

1. Use Toffu from Other AI Tools

Connect Toffu to Claude Desktop, Cursor, ChatGPT, and other MCP-compatible AI clients. This lets you access Toffu's marketing capabilities directly from your preferred AI assistant.

Available Tools

query_campaign_performance (Campaign Performance, All Platforms) - Campaign metrics across every connected ad platform (Google, Meta, LinkedIn), merged into one normalized table with per-currency totals. Read-only.

send_message (Ask Toffu) - Delegate any marketing task or question to Toffu, the AI marketing agent: analytics, content, reports, campaign work. Long tasks return a session id; collect the answer with get_task_result.

get_task_result (Get Toffu Task Result) - Collect the answer for a long-running send_message task. Read-only.

creative_report (Creative Performance Report, Meta Ads) - Every Meta ad with its creative (title, text, thumbnail) joined to performance: spend, CTR, CPC, conversions, CPA, ROAS. Read-only.

fetch_memory (Search Marketing Memory) - Search your company's marketing memory: strategy decisions, past campaigns, brand guidelines, learnings. Read-only.

propose_change (Propose Campaign Change) - Stage a campaign budget/status change and get a preview. Nothing executes at this stage.

apply_change (Apply Approved Change) - Execute a confirmed proposal. Disabled by default per workspace; all changes run under hard spend caps and are undoable.

undo_change (Undo Applied Change) - Revert an applied change to its pre-change values.

Budget changes are off by default for every workspace. Until an admin enables them, the change tools return a clear policy notice — by design.

Setup — One-Click Connect (Recommended)

Toffu supports OAuth sign-in, so most clients connect with no API keys and no config files:

  1. In your AI client, add a custom connector / MCP server with the URL:

    https://mcp.toffu.ai/mcp
    
    • Claude Desktop / claude.ai: Settings → Connectors → Add custom connector
    • ChatGPT: Settings → Connectors → Add connector (requires a plan with connector support)
  2. Click Connect. Your browser opens a Toffu authorization screen.

  3. Sign in to Toffu if needed, pick the workspace you want to grant access to, and click Approve.

  4. You're returned to your AI client with Toffu connected.

The client gets access to the workspace you approved, acting with your permissions. To revoke access later, disconnect the connector in your AI client.

Setup — API Key (for clients without OAuth)

Some clients (e.g. Cursor) connect with a static API key instead.

Step 1: Generate API Key

Go to Integrations → Scroll to MCP section → Generate API Key

Make sure to copy the API key when you generate it, it will not be shown again. If you lose it, you can generate a new one.

MCP API Key

MCP API Key

Step 2: Configure Your AI Client

Cursor

Add to your ~/.cursor/mcp.json:

{
  "mcpServers": {
    "Toffu": {
      "type": "streamable-http",
      "url": "https://mcp.toffu.ai/mcp",
      "headers": {
        "Authorization": "Bearer your-api-key-here"
      }
    }
  }
}

Replace your-api-key-here with your actual API key.

Start Using Toffu

Claude Desktop: Use @toffu fetch_memory "query" or @toffu send_message "request"

Cursor: Access Toffu tools through the MCP integration panel

Example Use Cases

Research: @toffu fetch_memory "competitor analysis" → Use insights for strategy

Content: @toffu send_message "Create LinkedIn post about our new feature"

Analysis: @toffu send_message "Analyze Google Ads performance this month"


2. Connect External Services to Toffu

Add MCP servers from external services (like Ahrefs, GitHub, databases, or custom tools) to Toffu. This expands Toffu's capabilities beyond its native integrations, allowing it to access additional data sources and tools when helping you.

Setup

  1. Go to AI Settings
  2. Scroll to the MCP Servers section
  3. Click Add MCP Server
  4. Enter the MCP server details (URL, authentication credentials)
  5. Save and enable the connection

Once connected, Toffu can automatically use these external services when relevant to your requests.

Popular MCP Servers

  • Ahrefs - SEO metrics and backlink analysis
  • GitHub - Repository data and code insights
  • PostgreSQL/MongoDB - Direct database access
  • Custom APIs - Your own internal tools

Troubleshooting

"Connection request expired" on the approval screen: The sign-in link is valid for a few minutes. Start the connection again from your AI client.

Stuck on Connect / sent back with an error: Make sure you're signed in to toffu.ai in the same browser, then retry. If your account has several workspaces, confirm you approved the one you meant.

Authentication errors (API key setup): Verify the API key is correct and hasn't been regenerated since you copied it.

Connection issues: Check network connectivity to mcp.toffu.ai

Tool errors: Ensure required integrations are enabled in your Toffu account

FAQ

What is the Toffu MCP server URL?

https://mcp.toffu.ai/mcp. Add it as a custom connector or MCP server in your AI client, then approve the workspace you want the client to reach.

How do I connect Toffu to Claude?

In Claude Desktop or claude.ai, open Settings → Connectors → Add custom connector, enter https://mcp.toffu.ai/mcp, and click Connect. Sign in to Toffu, pick the workspace, and approve. No API key is needed.

How do I connect Toffu to Cursor?

Cursor uses a static API key. Generate one in Toffu under Integrations → MCP → Generate API Key, then add the server to ~/.cursor/mcp.json with the key in an Authorization: Bearer header, as shown above.

Which tools does the Toffu MCP server expose?

Eight: query_campaign_performance, send_message, get_task_result, creative_report, fetch_memory, propose_change, apply_change, and undo_change. The first five are read-only; the change tools are off by default per workspace.

Can the MCP server change my ad campaigns?

Only if a workspace admin turns it on. Budget and status changes are disabled by default, run under hard spend caps, and are undoable with undo_change. Until enabled, those tools return a policy notice instead of acting.

Can Toffu use MCP servers from other services?

Yes. Add any external MCP server (Ahrefs, GitHub, a database, your own API) under AI Settings → MCP Servers, and Toffu will call it when it's relevant to your request.

Need Help?