Skip to main content
Glama
redactic-mcp

redactic-mcp

Official
by redactic

Redactic MCP server

Mask, restore and black out personal data in text and files from Claude Desktop, Claude Code, Gemini CLI, Cursor and any other MCP client.

The server is a thin client of the API of Redactic (redactic.eu). Your text and files are sent to Redactic servers in the European Union, where the personal data is detected. The AI model receives labels such as ⟦PERSON_1⟧. The map that turns the labels back into the original values stays in the memory of this server, on your computer.

What you need

  • A Redactic workspace and an API key. Create the key in Redactic under Admin, API. It starts with rk_live_ and is shown once. Every snippet below has rk_live_... where your key goes.

  • For every client except the Claude Desktop extension: Node.js 18 or later.

  • The folders in which the file tools may read and write, as full paths. Separate several folders with : on macOS and Linux and with ; on Windows. Inside a JSON file, write each Windows backslash twice.

System

Folder example

Inside a JSON file

macOS

/Users/you/Documents/redactic

"/Users/you/Documents/redactic"

Windows

C:\Users\you\Documents\redactic

"C:\\Users\\you\\Documents\\redactic"

Linux

/home/you/Documents/redactic

"/home/you/Documents/redactic"

Related MCP server: mahrem-mcp

Install

The same instructions are on redactic.eu/developers.

Claude Desktop

With the extension: download redactic-<version>.mcpb from redactic.eu/developers and compare its SHA-256 with the one printed on that page. Open the file, or drag it into Claude Desktop, or choose Settings, Extensions, Advanced settings, Install Extension. Claude Desktop asks for the API key, which it keeps in the keychain of your system, and for the folders the file tools may use. Claude Desktop brings its own Node.js, so nothing else needs to be installed.

With the configuration file, which needs Node.js: in Claude Desktop choose Settings, Developer, Edit Config. The file it opens is:

  • macOS: ~/Library/Application Support/Claude/claude_desktop_config.json

  • Windows: %APPDATA%\Claude\claude_desktop_config.json

  • Linux: ~/.config/Claude/claude_desktop_config.json

Add the server, save the file, then quit Claude Desktop completely and open it again.

{
  "mcpServers": {
    "redactic": {
      "command": "npx",
      "args": ["-y", "@redactic/mcp"],
      "env": {
        "REDACTIC_API_KEY": "rk_live_...",
        "REDACTIC_ALLOWED_DIRS": "/Users/you/Documents/redactic"
      }
    }
  }
}

On Windows, the folder line reads:

"REDACTIC_ALLOWED_DIRS": "C:\\Users\\you\\Documents\\redactic"

Claude Code

Run this command in a terminal. Keep the options before the name redactic: a name that comes right after --env is read as one more variable and the command fails. Everything after -- is the command that starts the server. With --scope user the server is available in all your projects.

claude mcp add --env REDACTIC_API_KEY=rk_live_... --env REDACTIC_ALLOWED_DIRS="$HOME/Documents/redactic" --transport stdio --scope user redactic -- npx -y @redactic/mcp

On Windows, write the folder as a Windows path:

claude mcp add --env REDACTIC_API_KEY=rk_live_... --env REDACTIC_ALLOWED_DIRS="C:\Users\you\Documents\redactic" --transport stdio --scope user redactic -- npx -y @redactic/mcp

A team can share the server in the .mcp.json file of a project and keep the key out of it. Each person sets REDACTIC_API_KEY in their own environment, and Claude Code reads it:

{
  "mcpServers": {
    "redactic": {
      "type": "stdio",
      "command": "npx",
      "args": ["-y", "@redactic/mcp"],
      "env": {
        "REDACTIC_API_KEY": "${REDACTIC_API_KEY}",
        "REDACTIC_ALLOWED_DIRS": "${REDACTIC_ALLOWED_DIRS}"
      }
    }
  }
}

Gemini CLI

Add the server to ~/.gemini/settings.json for all projects, or to .gemini/settings.json in one project.

Write the key in the env block. Gemini CLI removes from the environment it passes to a server every variable whose name contains KEY, TOKEN or SECRET, so a key set only in your shell does not reach the server.

{
  "mcpServers": {
    "redactic": {
      "command": "npx",
      "args": ["-y", "@redactic/mcp"],
      "env": {
        "REDACTIC_API_KEY": "rk_live_...",
        "REDACTIC_ALLOWED_DIRS": "/home/you/Documents/redactic"
      }
    }
  }
}

On Windows, the folder line reads:

"REDACTIC_ALLOWED_DIRS": "C:\\Users\\you\\Documents\\redactic"

Cursor

Add the server to ~/.cursor/mcp.json for all projects, or to .cursor/mcp.json in one project. Cursor can read the key from your environment with ${env:REDACTIC_API_KEY}.

{
  "mcpServers": {
    "redactic": {
      "command": "npx",
      "args": ["-y", "@redactic/mcp"],
      "env": {
        "REDACTIC_API_KEY": "${env:REDACTIC_API_KEY}",
        "REDACTIC_ALLOWED_DIRS": "/Users/you/Documents/redactic"
      }
    }
  }
}

The developers page also has an "Add to Cursor" link. It fills in the server with placeholder values: replace the key and the folder before you save.

ChatGPT, Microsoft 365 Copilot and claude.ai

This server runs on your computer, so an assistant in the browser cannot use it. ChatGPT and Microsoft 365 Copilot connect to remote MCP servers. A remote Redactic server is in preparation.

A file that you upload into a web chat reaches the model as it is. Mask it in Redactic first, then upload the masked copy.

Settings

Variable

Meaning

REDACTIC_API_KEY

Your API key. Required.

REDACTIC_ALLOWED_DIRS

The folders in which the file tools may read and write. Separate several folders with : on macOS and Linux and with ; on Windows. Folders can also be given as command line arguments after the name of the server, one argument each; the Claude Desktop extension passes them this way. Without any folder the file tools answer with the instruction to set one.

REDACTIC_REGIONS

Default countries of your documents, for example IT,DE (at most four). Without it Redactic detects the country.

REDACTIC_LANGUAGE

Default language hint, for example it.

REDACTIC_DETECTION_MODE

Default detection mode: accurate or fast.

REDACTIC_RESTORE_INLINE

Set to 1 to receive restored text in the conversation. The model then reads the original values.

REDACTIC_API_URL

Address of the API, for test environments.

Tools

Tool

What it does

mask_text

Masks a text and returns the masked text with a session_id. Pass the same session_id again and each person keeps the same label for the whole conversation.

restore_text

Puts the values back into a text with labels and writes the result to a new file. It returns the path of the file.

mask_file

Masks a file and saves <name> (masked) next to the original. It returns a session_id for restore_file.

restore_file

Puts the values back into a masked text, Word or Excel file and saves <name> (restored). It works on your computer and does not call the API.

redact_file

Blacks out a PDF or an image for good and saves <name> (redacted) with its certificate.

check_file

Checks a PDF or Word file for hidden data: text under shapes, metadata, older revisions, comments, attachments.

get_usage

Returns the usage of the month and the limits of the workspace.

Every masking tool accepts detection_mode: accurate (the default) takes a second look at places, addresses and organisations, and fast skips it and answers sooner.

How your data is handled

  • A text is processed in memory on Redactic servers in the EU and is gone when the answer is sent.

  • A file is deleted from the Redactic servers as soon as the server has saved the result on your computer.

  • The restore map is kept in the memory of this server for eight hours after its last use. It is never written to disk, never logged and never part of a tool result. When the server stops, the map is gone, so restore what you need before you close your MCP client.

  • restore_text writes to a file and returns its path, so the original values stay out of the conversation.

  • No tool overwrites a file. A new file gets a new name, with a number when the name is taken.

  • A masked PDF cannot be restored. Keep the original PDF when you may need its values again.

Limits

One document is processed at a time for each API key. When Redactic is busy the server waits and tries again for up to a minute, then tells you when to retry. The monthly allowance is the one of your Redactic plan; get_usage shows where you stand.

Development

npm install
npm test
npm run build:mcpb

The tests run against a fake API on localhost and need no key. npm run build:mcpb builds dist/redactic-<version>.mcpb and prints its SHA-256.

Licence

MIT, see LICENSE.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    Related MCP Connectors

    • Detect and redact PII and secrets before text reaches an LLM, with reversible placeholders.

    • Redact personal data from text before your agent sends it to a model or stores it: names, contact details, ID numbers, financial and health identifiers, card numbers, plus any exact terms you add. Nothing stored, no third-party AI. Tools - redact_text: returns the redacted text and a count of what was removed, by type. Never the removed values. Options: restrict to particular data types, add exact terms to remove, choose the replacement style (a type label such as <PERSON> by default) and set a confidence threshold. - list_entity_types: lists the detectable data types. Free. Data handling - Nothing you send is stored. Text is processed in memory and discarded when the call returns. - No third-party AI. Text is never sent to OpenAI, Anthropic, Google or any other outside model. - Tool arguments are never logged. Only counts are kept: credits used, a timestamp and how many of each type were found. PDFs, including scanned pages, are handled by the REST API with the same key and credits. They aren't offered as an MCP tool because a file doesn't fit in a tool call. Pricing: 1 credit per call plus 1 per 1,000 characters. 10 free credits on sign-up, no card. Credit packs from $5; purchased credits never expire. Failed calls and repeated identical calls are free. Setup and API key: https://eleyed.com/mcp?utm_source=glama&utm_medium=directory&utm_campaign=listing

      https://eleyed.com/api/mcp
    • EU descriptive AI redaction (GDPR & OSL: 98.3% precision, 93.2% recall). Zero retention.

      https://redact.aigine.se/mcp
    • Check a document for hidden text before your agent reads it. PDF, Office, RTF, HTML.

    Related MCP Servers

    • A
      license
      A
      quality
      B
      maintenance
      Enables AI agents to permanently remove personally identifiable information from PDF documents, returning redacted files with the sensitive text deleted rather than merely covered. Supports scanned documents, multiple languages, entity types, and optional human review.
      9
      782 npm
      1
      MIT
    • A
      license
      B
      quality
      C
      maintenance
      Enables AI assistants to mask personal data in local Turkish documents by accepting a local file path (TXT, MD, DOCX, or text PDF) and returning pseudonymized text such as [KISI-1] and [TCKN-1] instead of the original values. Tools scan for PII types, produce masked output, and restore real values into a local file without ever returning plaintext to the model.
      4
      1
      MIT
    • A
      license
      A
      quality
      B
      maintenance
      Enables an AI assistant to redact text and, more importantly, files without reading them, since file redaction and restoration happen locally and only counts are reported, keeping personal data out of the conversation. It exposes tools for redacting text, writing a redacted copy of a UTF-8 file with an owner-only restore mapping, and putting original values back offline.
      3
      MIT
    • F
      license
      Not graded
      quality
      F
      maintenance
      Enables Claude to work on documents containing personal data by pseudonymizing names, addresses, IBANs, health data and other GDPR categories locally before they are sent, then restoring the real values on your own machine. Supports Word, Excel, PowerPoint, PDF, CSV, Markdown, e-mail and other formats with consistent pseudonyms per project and no data leaving the computer.
      -