At a glance
| Item | Value |
|---|---|
| Type | Remote MCP server (Streamable HTTP), no local installation |
| Endpoint (Claude) | https://mcp.paperoffice.ai/claude |
| Other profiles | /chatgpt (alias /openai), /grok, /cursor, /dms |
| Authentication | OAuth 2.1 with PKCE (Claude, ChatGPT, Grok); bearer user/group tokens (po_ut_ / po_gt_) for Cursor |
Enterprise-managed authorization (ID-JAG / jwt-bearer) | not available yet |
| Tool annotations | title, readOnlyHint, destructiveHint on every tool |
| Processing location | PaperOffice EU infrastructure (own hardware) |
| Conversation data | not collected — only tool calls are logged |
| Operator | PaperOffice Enterprise Operations GmbH |
Domains to allow
mcp.paperoffice.ai— MCP endpoint, OAuth authorisation and token endpointsapi.paperoffice.ai— documentation (/latest/docs/llms.txt, Postman) and REST APIapp.paperoffice.ai— PaperOffice web app (token management, revocation, audit)
All traffic is HTTPS (TLS 1.2+). No inbound connection to your network is required.
Identity and access
- Each user connects with an individual PaperOffice login; permissions are the user's PaperOffice permissions (workspace roles, read-only, WORM, legal hold).
- Connections are visible and revocable per user under Account → API → MCP connections; account admins see all connections of the organisation.
- Token types
po_sk_andpo_pk_are rejected by the MCP server.
Logging and audit
Every tool call is recorded with timestamp, user, tool, parameters summary, credits and connection in the account audit trail (Account → API usage). Logs stay within PaperOffice.
Data protection documents
DPA generator, TOM overview and certifications: PaperOffice app → Trust Center. Privacy policy: paperoffice.ai.
Rollout recommendation
- Pilot with one workspace and read-only roles.
- Enable write roles per workspace after the pilot.
- Review API usage weekly during the first month.