The data flow in one picture
You (chat) → Assistant (Anthropic / OpenAI / xAI) → PaperOffice MCP server (EU) → your workspace
← tool result (only what was requested) ←- You write a request in the assistant.
- The assistant decides which PaperOffice tool to call and sends the parameters (e.g. a search term, a document ID) to
mcp.paperoffice.ai. - PaperOffice executes the call with your permissions in its own EU infrastructure and returns the result — for example the text of one document or a list of file names.
- The result becomes part of your conversation with the assistant and is handled under the assistant provider's terms.
What PaperOffice does not receive
PaperOffice does not receive your chat history, your prompts or the assistant's answers. Only the tool call itself (tool name, parameters, time, user) reaches the server. PaperOffice does not collect conversation data.
What PaperOffice logs
Every tool call is written to the account audit trail with time, user, tool, credits and connection. Administrators see it under Account → API usage. Document content is not duplicated into the log.
Where processing happens
PaperOffice processes documents, OCR and AI extraction on its own hardware in EU data centres. The MCP server is part of that infrastructure. No document is uploaded to the assistant provider by PaperOffice — the provider only receives what the assistant asked for and what you can already see in the chat.
Your controls
- Connect only the PaperOffice user whose workspaces should be visible.
- Use workspace roles to keep sensitive workspaces read-only or invisible.
- Revoke the connection at any time — see Disconnect Claude and revoke access.
- Retention rules (WORM), legal holds and the trash apply to connector actions exactly as in the app.
Documents
Privacy policy, data processing agreement (DPA generator) and technical and organisational measures are available in the PaperOffice app under Trust Center and at paperoffice.ai.