Zero-Knowledge MCP

The Zero-Knowledge MCP (Model Context Protocol) Server allows AI assistants (like Claude Desktop or Cursor) to execute tools that communicate with external APIs without ever exposing the API credentials to the AI agent or its context.

The Problem with Current MCP Servers

Most MCP servers configure their credentials via environment variables within the configuration file itself:

{
  "mcpServers": {
    "github": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": {
        "GITHUB_PERSONAL_ACCESS_TOKEN": "ghp_actual_token_here"
      }
    }
  }
}

When configured this way, the credential value is present in the configuration file, injected into process memory, and completely accessible to the AI model. A prompt injection attack or a compromised tool dependency could easily exfiltrate this key.

The Zero-Knowledge Approach

By routing requests through the AgentSecrets local proxy, the MCP server itself becomes completely credentialless.

{
  "mcpServers": {
    "zero-knowledge-mcp": {
      "command": "python",
      "args": ["/absolute/path/to/zero-knowledge-mcp/server.py"]
    }
  }
}

Notice the complete absence of an env block. The agent passes only a key name (e.g., GITHUB_TOKEN). The local AgentSecrets proxy resolves the actual value from the OS keychain and injects it at the transport layer before the request reaches GitHub.

Getting Started

1. Install AgentSecrets and log in:

   agentsecrets init

2. Clone the ZK MCP repository:

   git clone https://github.com/The-17/zero-knowledge-mcp
   cd zero-knowledge-mcp
   make install

3. Store your credentials:

   agentsecrets secrets set GITHUB_TOKEN=ghp_your_token

4. Allowlist the target domains:

   agentsecrets workspace allowlist add api.github.com

5. Start the Proxy:

   agentsecrets proxy start

You can now use Claude or Cursor to securely query GitHub without exposing your token to the LLM.

Building Custom Tools

Creating a zero-knowledge tool is as simple as defining the tool using the AgentSecrets Python SDK:

from agentsecrets import AgentSecrets

client = AgentSecrets()

async def fetch_github_repos():
    # The actual token is injected by the proxy. 
    # Your code only uses the reference key.
    response = await client.async_call(
        "https://api.github.com/user/repos",
        bearer="GITHUB_TOKEN"
    )
    return response.json()

For a full tutorial, check out theZero-Knowledge MCP Repository.