MCP Server: IDE Prompt Security Gateway for Cursor & Claude Desktop

Unblock Autonomous AI Coding: Zero Secret & PII Leaks in Cursor & Claude Desktop

Autonomous AI coding assistants—including Cursor Composer, Windsurf, and Claude Code—transform developer productivity by reading full repository trees, logs, and configuration files into LLM context windows. But this velocity introduces an immediate security risk: a simple prompt like "Fix database connection timeout" can cause an autonomous agent to read .env.production, database connection strings, and test databases containing real customer PII, transmitting them in plaintext to third-party model providers.

The PrivacyScrubber MCP Server gives engineering teams the ultimate outcome: unconstrained AI coding velocity with zero secret exfiltration risk. Operating directly on your machine over standard input/output (stdio), it intercepts file reads and prompt payloads locally, replacing secrets and personal data with reversible tokens ([API_KEY_1], [DB_URI_1], [NAME_1]) before data leaves your workstation.

+-----------------------------------------------------------------------------------+
| LOCAL WORKSTATION / IDE (Cursor, Claude Desktop, Windsurf, VS Code)               |
|                                                                                   |
|  1. Developer Prompt / Autonomous Agent File Read                                 |
|     "Analyze logs/orders.json and fix the failing payment transaction"            |
|           |                                                                       |
|           v (stdio JSON-RPC IPC - 0ms Network Latency)                            |
|  +-----------------------------------------------------------------------------+  |
|  | PRIVACYSCRUBBER MCP SERVER (Local Node.js Process)                          |  |
|  | - Intercepts sanitize_file / sanitize_text tool calls                       |  |
|  | - 30 Specialized Profiles (DevOps Secrets, AWS, DB URIs, GDPR, HIPAA)     |  |
|  | - Replaces sensitive values with [AWS_KEY_1], [DB_URI_1], [CUSTOMER_1]     |  |
|  | - Volatile In-Memory Session Map (Zero Disk Writes, Zero Remote Telemetry)  |  |
|  +-----------------------------------------------------------------------------+  |
|           |                                                                       |
|           v (Tokenized Prompt)                                                    |
|  2. Tokenized Payload Egress -> External AI API (Anthropic / OpenAI / DeepSeek)   |
+-----------------------------------------------------------------------------------+
                                    | (External Network)
                                    v
       External LLM analyzes code structure without learning actual secrets
                                    |
+-----------------------------------------------------------------------------------+
|  3. Model Response -> IDE Agent executes reveal_text                              |
|     Tokens rehydrated in local memory -> Developer sees working solution!         |
+-----------------------------------------------------------------------------------+

60-Second Setup: 1-Click Blueprints for Cursor & Claude Desktop

Deploy PrivacyScrubber into your AI IDE in under 60 seconds with zero network ports or firewall configurations:

{
  "mcpServers": {
    "privacyscrubber": {
      "command": "npx",
      "args": ["-y", "@privacyscrubber/mcp-server"],
      "env": {
        "PS_PROFILE": "devops",
        "PS_TIER": "pro"
      }
    }
  }
}

{
  "mcpServers": {
    "privacyscrubber": {
      "command": "npx",
      "args": ["-y", "@privacyscrubber/mcp-server"],
      "env": {
        "PS_PROFILE": "general",
        "PRIVACYSCRUBBER_TELEMETRY": "false"
      }
    }
  }
}

Developer Secrets & PII Coverage: Zero-Trust Detection for Source Code

Calibrated specifically for source code, database dumps, and infrastructure manifests:

Exposed MCP Tool Reference: Native Integration with Agent Context

Your AI coding assistant automatically discovers four native tools via the standard Model Context Protocol schema:

Security & Speed Comparison: Local MCP Gateway vs. Disruptive Cloud Proxies

How local stdio interception provides absolute security and uncompromised developer speed:

Enterprise Rule Governance: Enforce Zero-Leak Code Across 50+ Engineers at $99/mo Flat

In enterprise engineering teams, you cannot rely on individual developers configuring custom regex patterns manually. With the TEAMS Tier ($99/mo Flat), security architects push centralized detection schemas—such as proprietary account ID formats and internal microservice token standards—directly to every developer's MCP server through encrypted team sync.

For automated repository pre-audits prior to launching agentic workflows, explore the companion Automated Directory Auditing guide. If your team also works across web-based chat interfaces, pair your IDE gateway with our in-browser Chrome Extension.