Security and data

What happens to your team's sheets once they are in datatomcp: where they are kept, who can reach them, what AI agents can and can't do, what is recorded, and how to delete it. For the full list of providers and legal bases, see the privacy policy.

Where your data lives

• In the EU. Datasets and accounts are stored on our server in the European Union (hosted by Contabo). Encrypted backups go to Cloudflare R2 in Western Europe.
• One file per version. Each import or refresh is stored as a new version, so an answer always names the exact data it came from. Older versions are deleted on a schedule (the current one and the one before on Free; up to 20 from the last 30 days on Pro).
• Connected databases and apps. For PostgreSQL, MySQL, Stripe, HubSpot and Airtable, the credentials you enter are encrypted (AES-256-GCM), never shown again and never given to agents. Our server uses them only to read the table or object you picked, inside a read-only transaction for databases. We ask for read-only users and keys (for Stripe a restricted key, never the secret key), and we can't reach databases on private networks.
• Only the sheets you pick. Google access uses the drive.file scope: we can read the spreadsheets you choose in Google's picker, and nothing else in your Drive.

Who can reach it

• Workspaces and roles. Each team works in its own workspace. Owners manage people and settings, editors add data and approve definitions, viewers ask questions and read. Invites only work for the email address they were sent to.
• Private datasets. A dataset can be limited to named people inside the workspace.
• Agents per workspace. Each connected AI tool and each API key can be limited to some workspaces, so the agent you use for one client never sees another's data.
• No public links. Every request needs a sign-in, an OAuth token issued to a tool you approved, or an API key. You can see and revoke every connected tool and key in the app.

What agents can and can't do

• Read only. Agents ask through a few validated tools (list, describe, query rows, totals, distinct values). They can't change or delete your data, and they can't run raw SQL.
• Exact answers, computed by the database. Totals and counts are calculated by the database, not estimated by a model, and every answer names its dataset version.
• The team's definitions. When a definition like "net revenue" has been approved by a person, every agent uses it and the answer says who approved it.

What is recorded

• Every agent question. For each call: which agent, on behalf of whom, which dataset and version, the query it sent, rows returned, time taken and whether it worked. It is on the Audit page in the app and can be exported as CSV.
• Every change. The activity log lists who added, refreshed, linked or deleted what, and who invited whom.
• Every dashboard refresh. When, by whom, and from where (the app, the API or an agent).

AI inside datatomcp

• Never used for training. We don't train models on your data and don't sell it.
• Off until you turn it on. AI cleaning suggestions send a column profile and at most 50 sample rows to Mistral AI (processed in the EU), only when a workspace owner enables them and an editor asks. The in-app assistant runs on your own AI provider key, stored encrypted (AES-256-GCM) and never shown again.
• Your agents, your terms. The query results an agent asks for go to that agent's provider (for example Anthropic or OpenAI) under your account with them.

Deletion

• Delete a dataset or your whole account from the app: it is removed at once, and from backups within 14 days.
• Deleting an account also revokes every connected tool and API key, and our access to your Google account.

What we don't have yet

datatomcp is a small, independent product. There is no SOC 2 or ISO 27001 report and no single sign-on (SSO) yet. If your company needs a data processing agreement, a security questionnaire or a call about any of this, write to [email protected] and the founder answers.