Security
Connecting data to an AI should not mean handing over the keys
Here is exactly what happens to your credentials, who can use them, and what your assistant is — and is not — able to do.
Credentials
How provider credentials are handled
Encrypted at rest
Refresh tokens and API keys are sealed with AES-256-GCM before they reach the database, under a key the application never exposes.
Never in your browser
Credentials are decrypted on the server for a single provider call. No token is sent to frontend code or placed in a URL.
MCP tokens are hashed
An MCP token is shown once when you issue it. Only a hash and a short prefix are stored, so it cannot be read back.
Refreshed automatically
A background job renews tokens shortly before they expire. If a provider revokes access, the connection is flagged for reconnection.
Every access recorded
Credential use and MCP queries are logged with workspace, provider, and outcome, and query history is visible in the dashboard.
Revoke in one click
Disconnecting deletes the stored credential immediately and revokes it at the provider where the provider supports that.
Access control
Decide who can connect, query, and administer
Workspaces keep each business or client separate. Roles decide who can add connections and invite people, and single sign-on routes your domain through your identity provider.
Roles
Owner, admin, member, and viewer roles control who can manage connections, invite people, and change workspace settings.
Shared or personal connections
When you connect an account, choose whether teammates can query it.
SAML single sign-on
Add your IdP metadata to send everyone on your email domain through Okta, Entra ID, or another provider.
Per-client MCP tokens
Each assistant gets its own token, so you can see usage and revoke one client without affecting others.
Write actions, explained
Most of the 61 tools only read. 9 tools can change live data, and they are labelled as such to the assistant.
FAQ
Write actions
What can change, when, and how to keep it under control.
9 tools can: changing a Seller Central price, merchant-fulfilled quantity, or listing title; confirming a shipment; pausing, enabling, or budgeting Amazon Ads, Google Ads, and Meta campaigns; changing an Amazon Ads keyword bid; and responding to a Vendor Central purchase order.
Each write tool is described to the assistant as a WRITE operation for explicit requests, such as “pause campaign 42” or “set the price of SKU A to 49”. Most MCP clients also ask you to approve each tool call before it runs.
Yes. Write calls appear in the query log alongside reads, with the provider, status, and timing.
Configure your MCP client to require approval for every tool call, and only approve read tools. You can also issue separate MCP tokens for people who only need reporting.
Ask your first question in ten minutes
Create a workspace, connect one account, and add the MCP server to your assistant. No API keys to paste and nothing to export.