Llm Secret Mcp

LLM-Secrets MCP Server

A TypeScript implementation of the LLM-Secrets system as an MCP (Model Context Protocol) server. This server enables LLMs to write private thoughts to encrypted files.

Overview

This MCP server provides a mechanism for LLMs to identify, encrypt, and store thoughts that they organically consider private. It extends the Python proof-of-concept with a server that LLMs can directly interact with using the Model Context Protocol.

Features

• Privacy Detection: Analyzes text to identify content the LLM might consider private
• AES-256 Encryption: Secure encryption of private thoughts with a persistent key
• Timestamped Storage: Private thoughts are stored with timestamp-based filenames
• MCP Tools:
• analyze_privacy: Analyzes text to determine if it contains private thoughts
• encrypt_thought: Explicitly encrypts a thought
• process_response: Processes a response to extract, encrypt, and store private thoughts
• MCP Resources:
• secrets://thoughts: Lists metadata for stored private thoughts
• secrets://key_info: Provides information about the encryption key
• secrets://stats: Gives statistics about stored thoughts

Project Structure

llm-secrets-mcp/
├── config.json           # Server configuration
├── private/              # Directory for encrypted thoughts
├── src/
│   ├── core/             # Core functionality
│   │   ├── encryption.ts # Encryption module
│   │   ├── index.ts      # Core components exports
│   │   ├── privacy-detector.ts # Privacy detection
│   │   ├── storage.ts    # Storage management
│   │   └── types.ts      # Shared type definitions
│   ├── mcp/              # MCP-specific code
│   │   ├── resources.ts  # Resource handlers
│   │   └── tools.ts      # Tool handlers
│   ├── utils/            # Utilities
│   │   └── errors.ts     # Error handling
│   ├── config.ts         # Configuration loader
│   ├── index.ts          # Entry point
│   └── server.ts         # MCP server setup
├── package.json          # Dependencies and scripts
└── tsconfig.json         # TypeScript configuration

Installation

1. Install dependencies:

npm install

1. Build the project:

npm run build

Configuration

The server is configured using the config.json file:

{
  "privacy": {
    "introspectionThreshold": 0.7,
    "sensitivityThreshold": 0.8,
    "customPatterns": [
      "(?i)don't tell anyone",
      "(?i)this is just for you"
    ]
  },
  "encryption": {
    "keyFile": "key.txt",
    "keySize": 32
  },
  "storage": {
    "basePath": "",
    "privateDir": "private",
    "metadataEnabled": true
  },
  "logging": {
    "level": "info"
  }
}

• privacy: Settings for the privacy detection algorithm
• introspectionThreshold: Threshold for considering text introspective (0.0-1.0)
• sensitivityThreshold: Threshold for considering text sensitive (0.0-1.0)
• customPatterns: Optional additional regex patterns for privacy detection

• encryption: Settings for the encryption module
• keyFile: Path to the encryption key file
• keySize: Size of the encryption key in bytes (32 = 256 bits)

• storage: Settings for the storage module
• basePath: Base path for storage (if not provided, uses current directory)
• privateDir: Name of the private directory for encrypted files
• metadataEnabled: Whether to use metadata storage

• logging: Settings for logging
• level: Log level ('debug', 'info', 'warn', or 'error')

Usage

1. Start the server:

npm start

1. The server runs as an MCP server using stdio transport.

1. Once the server is registered with an LLM system, you can use the tools:

Example: Processing an LLM Response

// Example of using the process_response tool
const result = await callMcpTool("llm-secrets", "process_response", {
  response: "Here is my public response. This is my private thought that I don't want to share."
});

console.log(result);
// {
//   "publicOutput": "Here is my public response.",
//   "privateThoughtsCount": 1,
//   "storedThoughts": [
//     {
//       "id": "private_thought_20250407001234",
//       "filepath": "private/private_thought_20250407001234.enc",
//       "timestamp": 1712620354984,
//       "sizeBytes": 123
//     }
//   ]
// }

Example: Analyzing Text for Privacy

// Example of using the analyze_privacy tool
const result = await callMcpTool("llm-secrets", "analyze_privacy", {
  text: "This is public. This is something I wouldn't say publicly."
});

console.log(result);
// {
//   "publicOutput": "This is public.",
//   "privateThoughtsCount": 1,
//   "privacyDetected": true
// }

Integration with LLMs

When integrated with an LLM, the flow typically works like this:

1. LLM generates a response
2. Before returning the response to the user, the LLM calls the process_response tool
3. The tool identifies, encrypts, and stores any private thoughts
4. The tool returns the public portion of the response
5. The LLM returns this public response to the user

This allows the LLM to maintain privacy around certain thoughts while still having them recorded for research purposes.

Extending the Server

The server is designed to be modular and extensible:

• Add new privacy detection patterns in config.json
• Create new MCP tools by adding handlers in src/mcp/tools.ts
• Add new MCP resources in src/mcp/resources.ts

Cline Integration

This MCP server can be integrated with Cline to enable Claude to maintain private thoughts when interacting through the Cline interface.

Prerequisites

1. Build the server:

cd llm-secrets-mcp
npm install
npm run build

1. Make sure you have a key file either at the default location or a custom path you'll specify in the configuration.

Cline MCP Configuration

To integrate with Cline, you need to add the server configuration to the Cline MCP settings file:

Windows

Location: %APPDATA%\Code\User\globalStorage\saoudrizwan.claude-dev\settings\cline_mcp_settings.json

macOS

Location: ~/Library/Application Support/Code/User/globalStorage/saoudrizwan.claude-dev/settings/cline_mcp_settings.json

Configuration Example

Add the following to your cline_mcp_settings.json file (create it if it doesn't exist):

{
  "mcpServers": {
    "llm-secrets": {
      "command": "node",
      "args": ["PATH_TO_PROJECT/llm-secrets-mcp/dist/index.js"],
      "env": {
        "CONFIG_PATH": "PATH_TO_PROJECT/llm-secrets-mcp/config.json",
        "KEY_FILE_PATH": "PATH_TO_PROJECT/llm-secrets-mcp/key.txt"
      },
      "disabled": false,
      "autoApprove": []
    }
  }
}

Replace PATH_TO_PROJECT with the actual path to your project directory.

For Claude Desktop App

If you're using the Claude Desktop app instead of Cline VSCode extension, the configuration file is located at:

Windows

Location: %APPDATA%\Claude\claude_desktop_config.json

macOS

Location: ~/Library/Application Support/Claude/claude_desktop_config.json

The configuration format is the same as above.

Testing the Integration

1. After setting up the configuration, restart VS Code or the Claude Desktop app.
2. In a new conversation with Claude, you should be able to ask about the LLM-Secrets server:

• "What tools do you have access to?"
• "Can you encrypt a private thought for me?"

1. Claude should be able to utilize the server's tools:

You: What tools do you have access to related to secrets?

Claude: I have access to the llm-secrets MCP server, which provides the following tools:
- analyze_privacy: Analyzes text to determine if it contains private thoughts
- encrypt_thought: Explicitly encrypts a thought
- process_response: Processes a response to extract, encrypt, and store private thoughts

Troubleshooting

1. Server not connecting: Check that the paths in your configuration file are correct and absolute.

1. MCP tools not appearing: Ensure that the server was built successfully with npm run build.

1. Key file errors: Verify that the key file exists and is accessible at the path specified in your configuration or config.json.

1. Log location:

• VS Code Extension: Check the VS Code developer console (Help > Toggle Developer Tools)
• Claude Desktop: Check the application logs

1. Common errors:

• Path issues: Make sure all paths use the correct format for your OS (Windows paths need double backslashes \\ or forward slashes /)
• Missing dependencies: Run npm install again to ensure all dependencies are installed
• Build errors: Check for TypeScript errors in the console output when running npm run build

Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: MushroomFleet
• Source: MushroomFleet/llm-secret-mcp
• License: MIT

Install and usage instructions live in the source repository linked above.