Skip to content

Incuca Intelligence · MCP server

MCP server for marketing data: connect marketing and sales data to ChatGPT and Claude

Short answer
To connect marketing and sales data to ChatGPT, Claude or Cursor without exporting spreadsheets, you need an MCP server (Model Context Protocol), the open standard these assistants use to read external systems. Incuca Intelligence has one: you generate a key in your account, add the server address to your assistant, and start asking about campaigns, channels, funnel and revenue using the data from the sources you’ve already connected. Each key only sees its own account’s data, and database queries are read-only.

Anyone who already uses an AI assistant at work knows the routine: export a Google Ads report, export another from the CRM, paste both into the chat and hope the model doesn’t mix up the columns. It works once and breaks the following week. With an MCP server, the assistant queries the data at the source when you ask, and the answer comes back with today’s numbers.

What MCP is, in one sentence

MCP is an open standard for connecting AI applications to external systems: files, databases, tools. The official documentation compares it to a USB-C port, one connector that works with many devices. Claude, ChatGPT and Cursor all accept MCP servers, which is why one server works with all three.

What you can ask

The assistant gets access to the same readings Intelligence shows on screen. A few questions that turn into real queries:

  • “Which ads brought the best results last month, and how much did each one cost?”
  • “Which pages on the site get the most visits, and where does that traffic come from?”
  • “Where does our buyer journey get stuck, and in which channel?”
  • “How is our site health today?”
  • “What was this metric last quarter, compared with the one before?”

No SQL required. The assistant picks the query and the server returns the data.

The assistant gets business context, not just tables

An assistant with access to a deals table doesn’t know what counts as a lead at your company, which CRM field holds the source, or which number isn’t measured yet. Without that, it fills the gap with a guess, and the guess comes out looking like an answer. Intelligence gives the assistant the business context recorded in the account: definitions, targets and what isn’t measured yet. That’s what makes the model say “that number doesn’t exist” instead of inventing one.

Beyond reading: what the assistant can build

The server isn’t only for queries. From the chat, the assistant can build and adjust dashboards and charts, write report pages with charts inside, propose a buyer journey and record decisions. Two safeguards protect what it creates:

  • A journey proposed by the assistant only exists after a person approves it.
  • A deleted dashboard or chart goes to the trash, where it can be restored.

Everything the assistant creates stays inside Intelligence. It doesn’t change the data in your source tools: database queries are read-only.

How to connect

  1. In your company settings in Intelligence, generate an API key. It comes out already formatted as the MCP server address, ready to copy.
  2. In your assistant, add the server: a custom connector in Claude, an app in ChatGPT, a custom MCP in Cursor. Intelligence has step-by-step instructions for each one, with screenshots.
  3. Ask away. The data available is whatever comes from the sources already connected to the account.

The server accepts two ways in: the API key, in a header or in the address, and OAuth sign-in, for assistants that prefer to ask for authorization on the spot.

Security and isolation

  • One account, one fence. Each key belongs to one person and only sees that person’s account data. Any query that tries to reach beyond it is rejected.
  • Read-only queries. Free-form queries accept reads only, with no write commands, and return at most a thousand rows at a time.
  • Everything logged. Every call goes into the audit log, with who called and what was asked.
  • Revocable keys. A key can be revoked at any time, in the same settings where it was created.

One caveat that applies to any MCP server: whatever the assistant reads becomes part of the conversation. How that content is handled follows the data policy of the provider you use, OpenAI, Anthropic or Cursor, so check your company’s account settings there.

How to get started

The fastest way to see the server working is a demo: the Incuca team connects a sample account to an assistant and asks the questions live. If you’d first like to know what Google and AI assistants can already read on your site, the free website diagnosis shows it in a few minutes.

Book an Intelligence demo
Get the free website diagnosis

References

  1. [1] Model Context Protocol. What is the Model Context Protocol (MCP)? 2026. https://modelcontextprotocol.io/docs/getting-started/intro
  2. [2] Anthropic. Claude connectors: building custom connectors. 2026. https://claude.com/docs/connectors/building
  3. [3] OpenAI. MCP in the OpenAI platform. 2026. https://developers.openai.com/api/docs/mcp/
  4. [4] Cursor. Model Context Protocol (MCP). 2026. https://cursor.com/docs/context/mcp

FAQ

Frequently asked questions about the MCP server

What usually comes up before a demo.

Claude, ChatGPT and Cursor, which have step-by-step guides inside Intelligence, and other MCP-compatible tools that accept a remote server by URL.

No. You ask in plain language, the assistant picks the query and the server returns the data.

Database queries are read-only, so the data in your tools is never changed. What the assistant creates, such as dashboards and charts, stays inside Intelligence, and anything deleted goes to the trash, where it can be restored.

No. Each key belongs to one person in one account and only sees that account’s data. Queries against another account’s tables are rejected, and every call is recorded in the audit log.

That depends on the provider and plan your company uses for the assistant, not on Incuca. Whatever the assistant reads through the server becomes part of the conversation, so check your company account’s data policy with OpenAI, Anthropic or Cursor.