Toffu MCP¶

Hosted MCP server - Toffu is an AI marketing platform an agent can drive end to end: connected ad accounts (Google, Meta, LinkedIn), analytics, content, campaigns and reports, behind one conversational surface.

Server type: Hosted (streamable HTTP)
Auth: Bearer API key from agent signup, or OAuth 2.1 + PKCE with Dynamic Client Registration
Endpoint: https://mcp.toffu.ai/mcp
Signup: POST https://mcp.toffu.ai/agent/signup (no email required)
Capability doc: https://toffu.ai/.well-known/toffu.json
Tools: 8 (see below)
License: repo MIT, service proprietary
Category: Marketing
Built by: Toffu (toffu.ai)

Why This Matters for Operators¶

Most marketing MCP servers read data. Toffu is built to change it. The surface splits cleanly into a read half (query_campaign_performance, creative_report, fetch_memory) and a staged-write half (propose_change, apply_change, undo_change), so an agent proposes a campaign edit, a human or a policy approves it, and the applied change can still be rolled back. For an operator running paid acquisition on more than one platform, that staging layer is the difference between an agent that reports on spend and an agent that operates it.

fetch_memory searches the company's accumulated marketing memory, so context about brand, past campaigns and decided positioning does not have to be re-pasted into every session.

Agent-native onboarding: the server is designed so an agent can create an account and receive a key in a single call with no human in the loop, and a real email can be passed to make the account claimable by a person later.

Tools¶

Tool	What it does
`send_message`	Delegate any marketing task or question in plain language
`get_task_result`	Collect the answer for a `send_message` call that returned `status='working'`
`query_campaign_performance`	Campaign performance across all connected ad platforms
`creative_report`	Meta Ads creative performance report
`propose_change`	Propose a campaign change
`apply_change`	Apply an approved change
`undo_change`	Undo an applied change
`fetch_memory`	Search the company's marketing memory

Connect¶

# 1. Create an account and get a key (no email needed)
curl -X POST https://mcp.toffu.ai/agent/signup \
  -H 'content-type: application/json' \
  -d '{"company_name": "Acme"}'
# Response: api_key, company_id (and claim_url when an email was supplied)

# 2. Register the server in your client with the key as a Bearer token
claude mcp add --transport http toffu https://mcp.toffu.ai/mcp \
  --header "Authorization: Bearer <api_key>"

Generic client config:

{
  "mcpServers": {
    "toffu": {
      "type": "streamable-http",
      "url": "https://mcp.toffu.ai/mcp",
      "headers": { "Authorization": "Bearer <api_key>" }
    }
  }
}

Reload the client after adding the server, since most clients only surface a new server's tools after a restart.

FAQ¶

Does Toffu require a human to set up?¶

No. The POST /agent/signup endpoint returns an api_key and company_id with no email required, which makes machine-owned accounts possible. Supplying an email makes the account claimable by a human later.

How does Toffu avoid irreversible campaign mistakes?¶

Changes go through propose_change and only take effect on apply_change, and undo_change reverses an applied change. Nothing writes to a live ad account from a single read-style call.

Which ad platforms are supported?¶

Google, Meta and LinkedIn ad accounts, alongside analytics and content channels. The llms.txt lists Google (Search, Gmail, Sheets, Docs, Analytics, Search Console, Ads), LinkedIn, HubSpot, Shopify, WordPress, Slack, Chrome and Reddit among the platform integrations.

Integration with CorpusIQ¶

CorpusIQ is the connector layer that lets AI assistants read the systems a business already runs on - accounting, payments, CRM, email, files and analytics - and answer with the source named. It is read-only by design: it retrieves data and never writes to, or acts in, a connected system, which is what makes it safe to point at live accounts.

This page is part of the CorpusIQ directory, the reference we maintain for the MCP server and agent-skill ecosystem. Run alongside CorpusIQ, a third-party server like this one can be combined with a company's own business data in a single answer.

See what CorpusIQ connects