Skip to content
MCP server

PrivacyFence

Approval gateway that keeps AI assistants from reading or changing your data unchecked

Runs locallyStatus and history
Install
Pricing
Open source
Platforms
macOS, Windows, Linux
Runs
Locally
Version
7.0.2
Listing updated
3 days ago

About PrivacyFence

PrivacyFence is an open-source MCP server for security. It runs locally on your machine and is started by your AI client.

PrivacyFence sits between an MCP client and business systems such as Gmail, Google Drive, Calendar, Slack, Salesforce, Jira, Confluence and Telegram. The assistant asks, PrivacyFence decides, and sensitive reads and consequential writes appear as approval cards showing the real content or change, who is asking and why, instead of a raw tool name or JSON.

Routine requests can be automated with narrow always-allow rules, such as a sender domain, a Drive folder, a Slack channel or a Jira project. Optional local PII detection flags likely personal data before it reaches the AI and can force a card even when a rule would allow it. Every decision goes into a chained audit log that detects edits made without its key. It is a defined set of connectors with each tool gated in code, not a generic proxy for arbitrary MCP tools.

Credentials stay with PrivacyFence, which runs under its own service account so the client cannot read them or approve its own requests, and no data passes through a PrivacyFence-operated server. The recommended single-user setup is now a Docker image on localhost with a passkey. Desktop installers for macOS, Windows and Linux are deprecated in version 7, and version 8 removes desktop mode. An organization deployment lets claude.ai, ChatGPT and Gemini Enterprise connect through OAuth.

Install

Run in a terminal

claude mcp add privacyfence -- npx -y https://github.com/privacyfence/privacyfence/releases/download/v7.0.2/PrivacyFence.mcpb

Add to claude_desktop_config.json

{
  "mcpServers": {
    "privacyfence": {
      "command": "npx",
      "args": [
        "-y",
        "https://github.com/privacyfence/privacyfence/releases/download/v7.0.2/PrivacyFence.mcpb"
      ]
    }
  }
}
Install in CursorOpens Cursor, which asks you to confirm. Or add it by hand:

Add to ~/.cursor/mcp.json

{
  "mcpServers": {
    "privacyfence": {
      "command": "npx",
      "args": [
        "-y",
        "https://github.com/privacyfence/privacyfence/releases/download/v7.0.2/PrivacyFence.mcpb"
      ]
    }
  }
}
Install in VS CodeOpens VS Code, which asks you to confirm. Or add it by hand:

Add to .vscode/mcp.json

{
  "servers": {
    "privacyfence": {
      "type": "stdio",
      "command": "npx",
      "args": [
        "-y",
        "https://github.com/privacyfence/privacyfence/releases/download/v7.0.2/PrivacyFence.mcpb"
      ]
    }
  }
}

Add to ~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "privacyfence": {
      "command": "npx",
      "args": [
        "-y",
        "https://github.com/privacyfence/privacyfence/releases/download/v7.0.2/PrivacyFence.mcpb"
      ]
    }
  }
}

Add to your client's MCP settings (most use this shape)

{
  "mcpServers": {
    "privacyfence": {
      "command": "npx",
      "args": [
        "-y",
        "https://github.com/privacyfence/privacyfence/releases/download/v7.0.2/PrivacyFence.mcpb"
      ]
    }
  }
}

What it does

Human approval cards
Sensitive reads and writes show the actual content before they proceed.
Policy rules
Narrow always-allow rules let routine requests run without a prompt.
Local PII detection
Highlights likely personal data before it reaches the AI.
Chained audit log
Records accepted, denied and automatic decisions and detects tampering.
Separate credentials
Connector sign-ins live under PrivacyFence's own service account.
Docker or organization mode
Runs on localhost for one person or as a central server for a team.

Good for

  • Let Claude read mail while approving sensitive messages
  • Limit an assistant to one Jira project or Drive folder
  • Keep an audit trail of AI access to company systems
  • Stop an agent approving its own requests
  • Roll out governed AI access across an organization

Pricing

Open source

Prices change. Check privacyfence.eu before you buy.

PrivacyFence FAQ

Does data pass through a PrivacyFence server?

No. The README says no data passes through a PrivacyFence-operated server.

What licence is it?

Apache 2.0, free and open source.

What is the recommended install?

The Docker image on your computer in single-user mode, with a passkey; desktop installers are deprecated.

Can it proxy any MCP server?

No. It offers a defined set of connectors with fixed gates, not a generic proxy.

Is PrivacyFence free?

Yes. PrivacyFence is open source, so you can use and self-host it for free.

How do I add PrivacyFence to Claude Code?

Run this command in your terminal: claude mcp add privacyfence -- npx -y https://github.com/privacyfence/privacyfence/releases/download/v7.0.2/PrivacyFence.mcpb

What are the best alternatives to PrivacyFence?

The closest alternatives to PrivacyFence are REA, Ha MCP, Concord MCP, Agent Roast and Midplane.

Alternatives to PrivacyFence

  1. REA

    MCP server

    CLI and MCP server that lets coding agents reverse engineer apps and binaries

    Runs locally
    • Open source
    • 464K GitHub stars
    • Developer tools
  2. Ha MCP

    MCP server

    MCP server for managing and automating Home Assistant from AI assistants

    Runs locally
    • 8 tools
    • Open source
    • 7.9K GitHub stars
    • Productivity
  3. Concord MCP

    MCP server

    Local-first MCP server for messaging and shared work-state between coding agents

    Runs locally
    • Freemium
    • 1.3K GitHub stars
    • Developer tools
  4. Agent Roast

    MCP server

    Offline review that flags agent changes which weaken tests, assertions or CI

    Runs locally
    • Open source
    • 4 Hacker News points
    • Developer tools
  5. Midplane

    MCP server

    Postgres gateway for AI agents with table grants, column masking and approval for writes

    Runs locally
    • Open source
    • 5 GitHub stars
    • Databases
  6. Leash

    MCP server

    Review what your AI touched, approve risky actions and undo supported changes

    • 1 tool
    • Remote
    • Free
    • 1 GitHub star
    • Security