AI Connector (MCP Integration)

An AI connector, such as one built on the Model Context Protocol (MCP), is a link that lets an AI assistant read data from or take actions in one of your business apps, using a user's permissions.

What is an AI connector?

An AI connector is a link that lets an AI assistant reach into one of your business systems, such as a CRM, help desk or file store, to read data or take actions there. You might ask an assistant to "find all open deals over $50,000," and the connector fetches the records for it.

Some connectors use the Model Context Protocol (MCP), a standard way for AI assistants to talk to other software. A vendor that offers an "MCP server" is offering a door that outside AI assistants can use to work with your account.

Insightly's Terms of Service are a clear example. Since October 14, 2026, they define a "Connected AI Assistant" as any outside AI assistant, agent or app that you or a user connects through its MCP Server, as Venpo reported.

How data moves through a connector

  1. A user links an AI assistant to a business app and approves access.
  2. The user asks the assistant a question or gives it a task.
  3. The assistant calls the app's connector, using that user's login and permissions.
  4. The app sends back records or carries out the action.
  5. The assistant's provider processes that data under its own terms.

Step five is easy to miss. Once data leaves the app, the app's contract may no longer cover it.

What it looks like in a vendor's terms

A typical clause reads like this:

You may connect third-party AI assistants to the Services through our connector. A connected assistant acts with the permissions of the user who authorizes it, and its activity counts as that user's activity. Connected assistants are third-party services, not our subprocessors. Data they retrieve is disclosed at your direction and handled under the assistant provider's terms.
  • "Permissions of the user" means the assistant can see and do whatever that person can. A user with admin rights gives the assistant admin reach.
  • "Counts as that user's activity" makes you responsible for what the assistant does, including mistakes.
  • "Not our subprocessors" means the assistant provider will not appear on the app's subprocessor list, and the app's DPA does not cover it.
  • "Disclosed at your direction" means the vendor treats you as the one sending the data out.

Insightly's terms use this structure, stating that each connected assistant "is not a service provider or subprocessor of Insightly."

A real change we caught

Venpo flagged this change: Insightly's new AI connector is not covered by its SOC 2 report. Insightly's privacy policy now says its MCP Server launched after its last audit window and "is not within the scope of the current report." A SOC 2 report that predates a feature tells you nothing about how that feature is secured.

Connectors can also come with new responsibility rules. Venpo also reported that OpenAI's app terms now cover ChatGPT calling your app on its own.

Why it matters to you

Data can leave through a side door. A connector can move customer records from a system you vetted into an AI tool you never reviewed. That AI tool may keep or use data under different rules.

Your audit evidence may have a gap. A new connector can ship between audits, as Insightly's did. Check the SOC 2 scope before telling a customer that a vendor's SOC 2 covers everything.

You answer for what the assistant does. If an assistant deletes records, emails the wrong people or changes data, terms like Insightly's treat that as your own action.

Data protection promises may not follow the data. If your own customers' personal data flows out through a connector, you may need to list or assess the AI provider in your own records and contracts.

Not every safeguard applies. Vercel's AI Gateway documentation warns that tool calls "may not be covered" by a provider's zero data retention policy. Connector traffic can fall into the same gap.

What to check

  • List which apps offer connectors or MCP servers, and which AI assistants your team has linked to them.
  • Read the app's terms for who is responsible for assistant activity and whether the assistant counts as a subprocessor.
  • Check whether the connector is inside the vendor's SOC 2 or other audit scope.
  • Limit connectors to users with the least access needed, and avoid linking admin accounts.
  • Find out how to revoke a connector immediately, and test that it actually cuts access.
  • Review the AI assistant provider's own terms for training and retention of connected data.

How Venpo helps

Venpo watches your vendors' terms, privacy policies and security pages, and emails you when a connector or MCP feature is added with new responsibility, data sharing or audit scope language. Each vendor's change history shows exactly what was added.

Start for free

Frequently asked questions

What is an MCP server in a SaaS product?

It is a connection point the vendor provides so outside AI assistants can work with your account in that product. The assistant uses a user's login and permissions to read data or take actions.

Is an AI assistant connected through MCP a subprocessor?

Some vendors say no, because you chose to send the data to the assistant. In that case the app's DPA does not cover what the assistant provider does with it.

How do I stop an AI connector from accessing my data?

Disconnect it in the assistant's settings and in the app's admin panel, then confirm access is gone. Insightly's terms, for example, warn that disconnecting on the assistant side "may not immediately revoke the underlying authorization."

Related terms

Sources

Last reviewed