# HLOS Documentation: MCP Integration

> Source: https://hlos.ai/docs/mcp
> Last updated: 2026-01-10
> Context: HLOS is a financial control plane for AI agents. Agents never receive credentials directly.

# MCP Integration

AI-powered secrets management for Cursor, Claude, and more.

The Model Context Protocol (MCP) allows AI assistants to help you manage secrets directly from your IDE—without ever exposing secret values to the AI.

## What is MCP?

MCP (Model Context Protocol) is an open standard that allows AI models to securely interact with external tools and data sources. When you connect HLOS via MCP, your AI coding assistant can list secrets, suggest missing ones, and even queue new secrets for approval—all without seeing the actual secret values.

### You Host the Agent • We Handle Identity

HLOS is an **identity and credential layer**, not an agent hosting platform. You deploy and run your agent (Claude, GPT, Gemini, etc.) on your own infrastructure. Your agent calls HLOS via MCP when it needs credentials, capabilities, or identity verification.

### Security First

The HLOS MCP server is designed with a **zero-value exposure** model. The AI can see secret names, providers, and metadata—but **never the actual values**. Any actions that modify secrets require your explicit approval in the HLOS dashboard.

## IDE Setup

### Cursor

AI-first code editor.

1. Open Cursor Settings → Features → MCP Servers
2. Or edit `~/.cursor/mcp.json` directly
3. Add the HLOS configuration below and restart Cursor

```json
{
  "mcpServers": {
    "hlos": {
      "command": "npx",
      "args": ["-y", "@hlos/mcp-server"],
      "env": {
        "HLOS_API_URL": "https://api.hlos.ai"
      }
    }
  }
}
```

### Claude Desktop

Anthropic's desktop app.

1. Open your Claude Desktop config file
2. **macOS:** `~/Library/Application Support/Claude/claude_desktop_config.json`
3. Add the HLOS configuration and restart Claude

```json
{
  "mcpServers": {
    "hlos": {
      "command": "npx",
      "args": ["-y", "@hlos/mcp-server"]
    }
  }
}
```

### Other IDEs

Any IDE or tool that supports MCP can use HLOS. The server is available via:

`npx -y @hlos/mcp-server`

Windsurf, Continue.dev, Zed, and Google's AI tools are all compatible.

## Available MCP Tools

*   `list_secrets`: List all secrets in a space (names and metadata only, never values).
    *   Use case: Understand what secrets exist before performing operations.
*   `check_secret_exists`: Check if a specific secret exists in a space.
    *   Use case: Verify a secret is available before using it, or avoid duplicates.
*   `get_secret_info`: Get detailed metadata about a specific secret (not the value).
    *   Use case: Check rotation status, provider type, and associated health issues.
*   `suggest_missing_secrets`: Analyze code context and suggest which secrets might be needed.
    *   Use case: Identify required API keys from code patterns automatically.
*   `create_secret`: Queue a secret creation request (requires user approval).
    *   Use case: Let AI suggest and queue new secrets without exposing values.
*   `check_health`: Check health status of secrets (rotation, exposure, etc.).
    *   Use case: Identify outdated or potentially compromised secrets.

## Learn More

*   [Security Model](/docs/security): How we keep secrets safe
*   [Provider Templates](/docs/providers): 21+ pre-configured providers