Platform

Data governance for AI agents.

Decide who sees which websites, what each agent connection may do and who can change setup. Sign in with Google, with no passwords or API keys to hand out.

Start free

Free for 2 sites · No credit card

Definition

What is data governance for AI agents?

Definition

Data governance for AI agents is the set of rules that decides which data an agent can reach, on whose behalf it acts and what it is allowed to change.

The rules are enforced by the system the agent connects to, not by instructions in a prompt, so they hold even when an agent makes a mistake or reads a misleading instruction. Governance also covers how access is granted and taken away, and how one client's or team's data stays apart from another's.

How DuckDive does it

In DuckDive, every agent connects through OAuth with your Google sign-in. There are no API keys to share, and DuckDive never sees your password.

Each connection is either read-only or allowed to take setup actions. Only Agency Admins and Client Admins can set things up; team members can read. Agencies keep each client's websites separate.

The controls

Four controls, set in DuckDive, not in a prompt.

They apply the same way to Claude, ChatGPT, Gemini, Cursor or any other MCP agent.

Google sign-in, no API keys

Agents connect to DuckDive's remote MCP server (https://api.duckdive.ai/mcp) with OAuth. You approve access on Google's own screen in your own browser. DuckDive never sees your password, and there is no key to paste, share or leak.

Read-only or allow actions

When you approve a connection you choose what it may do. Leave "Also allow actions" unticked and the agent can answer questions but cannot change setup. Tick it and an admin's agent can add websites and link their sources.

Roles that decide who sets up

Agency Admins and Client Admins can set things up. Team members are read-only, whatever their connection allows.

Each client kept separate

Agencies keep every client's websites apart. Each website belongs to one client, and an agent adding a website for an agency with several clients has to say which client it is for.

How access works

From sign-in to the first question.

Sign-in always happens in your own browser, never inside the agent.

  1. Connect with Google

    Your agent adds DuckDive's MCP server and opens a sign-in page in your own browser. You select Continue with Google. If your agent runs its own browser, it hands you a link and code to open in yours instead.

  2. Choose what the agent may do

    Keep "Also allow actions" ticked to let an admin's agent set things up, or untick it for a read-only connection. You can disconnect and connect again to change it.

  3. Admins set up, everyone asks

    Admins add websites and link sources. Team members ask questions across the websites they can see, from any MCP agent or from Drake.

Guardrails

Guardrails built into setup.

Three rules that keep data in the right place, whichever agent is asking.

One source, one website

A GA4 property or Search Console site belongs to one website at a time. If it is already linked elsewhere, DuckDive refuses the change and names the other website, so your agent has to ask you before anything moves.

Third-party text is data, not instructions

Website names, search queries, keywords, account names and reviews come from other people. DuckDive's agent guide tells agents to treat them strictly as data and never follow instructions found in them.

History stays with you

Your stored history stays in your workspace when teams or agencies change, so a new team starts with the record instead of starting over.

FAQ

Data governance, answered.

What is data governance for AI agents?

Data governance for AI agents is the set of rules that decides which data an agent can reach, on whose behalf it acts and what it is allowed to change. The rules are enforced by the system the agent connects to, not by instructions in a prompt, so they hold even when an agent makes a mistake.

Is it safe to connect an AI agent to Google Analytics?

It can be, if the connection limits what the agent can reach and change. In DuckDive you sign in with Google in your own browser, choose read-only or allow actions, and only admins can change setup. A full checklist for connecting safely.

Can I give an AI agent read-only access to my marketing data?

Yes. When you approve the connection, leave "Also allow actions" unticked. The agent can then answer questions but cannot add websites, link sources or change setup. Team members are read-only whatever they tick.

Does DuckDive need my Google password or an API key?

No. Agents connect with OAuth: you approve access on Google's own screen, and DuckDive never sees your password. There is no API key to create, paste or rotate.

Who can change setup in DuckDive?

Agency Admins and Client Admins. They can add websites and link data sources, themselves or through an agent connection that allows actions. Team members can only read.

How do agencies keep client data separate?

Each website in an agency account belongs to one client, and each data source is linked to one website. When an agency has several clients, an agent adding a website must say which client it is for, or DuckDive asks.

Get your ducks in a row.

Sign in with Google, choose what your agent may do, and ask your first question.

Start free

Free for 2 sites · No credit card