Integrations

Safe-Doc fits your tools - no copy-paste required.

Pseudonymization callable via REST API, MCP server (Model Context Protocol), and the new Safe-Doc skill for Claude Cowork. Same engine as the web UI: 90+ entities, zero storage.

NEW

New - Safe-Doc inside Claude Cowork

Pseudonymize your documents without leaving Cowork: Claude only works on pseudonymized data, and the restored file lands back on your machine, next to the original.

How it works

Install the Safe-Doc skill in Claude Cowork.
Point to a document from Cowork - it is processed by Safe-Doc (EU servers, self-hosted models) and returns pseudonymized.
Work in Cowork on the pseudonymized version - Claude only sees pseudonymized data.
Restore on your machine via the API: the result is saved on your workstation, next to the original file.

Why it's safe

Claude (Cowork) never receives your real data - only the pseudonymized version. Real values only pass through two places: your workstation, and our European servers (in-memory processing, ephemeral, self-hosted models, zero third-party AI). Restoration happens locally, next to the original - no lock-in.

« Zero third-party AI » applies to our processing engine: no external AI provider sees your content during pseudonymization. You then use Cowork - and Claude only sees pseudonymized data.

Get the Safe-Doc skill Join the Beta

Skill distribution: contact us to receive the installation package and Cowork setup guide. Full user guide →

Safe-Doc REST API

Automate pseudonymization before ChatGPT, Claude, Gemini or any internal pipeline. Ideal for scripts, orchestrators (Make, n8n), legal CRMs or line-of-business portals.

Use cases

Anonymize before ChatGPT via API: pseudonymize an excerpt or file before sending to a third-party LLM.
Batch processing: embed Safe-Doc in a document workflow (upload → pseudonymization → review → export).
Data Room: consistent pseudonyms across multiple files via the jobs API.
Restoration: retrieve the JSON mapping to de-pseudonymize AI output locally.

Documentation: app.safe-doc.ai/docs (OpenAPI, Bearer auth, curl examples).

Plans including API access: Cabinet+ and Enterprise. API key generated from your account dashboard.

Sample request

POST https://app.safe-doc.ai/v1/pseudonymize
Authorization: Bearer sk_live_…
Content-Type: application/json

{
  "text": "The agreement between John Smith and Morrison Partners sets a fee of £14,500.",
  "mode": "pseudonymize",
  "locale": "en"
}

Sample response

{
  "pseudonymized_text": "The agreement between [PERSON_1] and [ORG_1] sets a fee of [AMOUNT_1].",
  "entities_detected": 3,
  "mapping": {
    "PERSON_1": "John Smith",
    "ORG_1": "Morrison Partners",
    "AMOUNT_1": "£14,500"
  },
  "residual_risk_score": 0.12
}

For PDF/DOCX files, use the async jobs endpoint (POST /jobs): create, presigned upload, status, download. The download returns the pseudonymized text + mapping (JSON) - not a reconstructed same-format file. See full documentation.

MCP - recommended

Safe-Doc MCP server - plug pseudonymization directly into Claude Cowork and any MCP-compatible client.

The Model Context Protocol is an open standard: your AI assistant calls Safe-Doc as a native tool, with no manual copy-paste.

Why MCP?

Native workflow: anonymization becomes an automatic step before AI analysis.
Consistent pseudonyms: the same entity keeps the same token across every file in a Data Room.
Zero storage: ephemeral processing, no document content retained.
Open standard: compatible with Claude Desktop, Claude Cowork, Cursor and any MCP client - no vendor lock-in.

Server: @safedoc/mcp (npm) · API base URL: https://app.safe-doc.ai

Connection (Claude Desktop / Cowork)

Add to your MCP configuration (claude_desktop_config.json or Cowork equivalent):

{
  "mcpServers": {
    "safedoc": {
      "command": "npx",
      "args": ["-y", "@safedoc/mcp"],
      "env": {
        "SAFEDOC_API_KEY": "sk_live_…"
      }
    }
  }
}

Exposed tools

safedoc_pseudonymize_text - pseudonymize a text excerpt (90+ PII entity types).
safedoc_pseudonymize_document - extract and pseudonymize the text of a PDF or DOCX file; returns pseudonymized text + mapping (the original file is not reconstructed).
safedoc_restore - re-inject original values into AI output via mapping.
safedoc_risk_report - residual risk score and detected entity list.

Plans: MCP available on Cabinet+ and Enterprise, same API key as the REST endpoint.

Open standard vs proprietary integration

Safe-Doc remains a protection layer: you keep your AI tool. Unlike "workspace + integrated chat" platforms that impose their ecosystem, API and MCP let you choose the model and client.

	Proprietary integration	Safe-Doc (open API + MCP)
Access	Single-vendor plugin or extension	REST API + MCP: any compatible client
Your documents	Often stored in third-party workspace	Never retained after processing
Your AI queries	Transit through the platform	Safe-Doc never sees the query
Model choice	Locked to the vendor	ChatGPT, Claude, Gemini, Copilot - your choice
Automation	Workflow imposed by the tool	Copy-paste, API or MCP - at your pace
Claude Cowork	Proprietary channel lock-in	Safe-Doc skill + open MCP - pseudonymize before Claude sees your data
Lock-in	Dependency on closed ecosystem	Open standard, model-agnostic layer

Get started

View documentation

Join the Beta

Technical questions or Enterprise deployment? Contact us.