AgentSecrets

What it does

AgentSecrets provides a secure, zero-knowledge infrastructure for managing AI agent credentials. Instead of agents retrieving raw secret values into memory—where they can be leaked via prompt injection or logs—AgentSecrets resolves credentials outside the agent and injects them directly at the transport layer.

Tools

• call: Execute API requests with credentials injected via bearer tokens, custom headers, or query params without the agent seeing the secret.
• proxy: Start a zero-trust HTTP proxy that enforces domain allowlists and redact response bodies.
• secrets: Manage, sync, and set per-secret policies (e.g., requiring manual approval for POST requests).
• agent: Manage agent identities and issue cryptographically verified tokens with specific capability bounds.
• env: Inject secrets into child process environment variables at spawn time with stdout redaction.

Installation

To install the MCP server for Claude Desktop or Cursor, run:

agentsecrets mcp install

In your claude_desktop_config.json:

{
  "mcpServers": {
    "agentsecrets": {
      "command": "/usr/local/bin/agentsecrets",
      "args": ["mcp", "serve"]
    }
  }
}

Supported hosts

• claude
• cursor
• codex