> ## Documentation Index
> Fetch the complete documentation index at: https://docs.karada.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Playbook: Building an AI Agent

> A step-by-step playbook on using Karada to connect an AI Agent to your APIs.

Karada enables LLM-powered agents to safely interact with your proprietary systems. This playbook walks you through building an AI agent that uses a Karada-deployed MCP server to fetch data and perform actions on your behalf.

## Prerequisites

* An existing REST API with an OpenAPI (v3) or Swagger specification.
* A Karada account.
* An AI client that supports the Model Context Protocol (e.g., Claude Desktop, Cursor).

***

## Phase 1: Ingesting your API

To let the AI understand your system, you must expose your API logic as discrete tools. Karada handles this automatically.

1. Navigate to the **Karada Dashboard** and create a new **Auto-MCP** project.
2. Provide your OpenAPI specification.
   * *Tip*: If you don't have a spec file ready, you can paste a URL to your public `openapi.json` endpoint.
3. Karada immediately generates a ready-to-deploy MCP server. Each endpoint from your spec is converted into a structured `tool` with a detailed description derived from your schema.

## Phase 2: Deploying the Server

Your generated server needs to run continuously so the agent can ping it anytime.

1. Click **Deploy** from your project page.
2. Karada provisions a secure, isolated microVM sandbox and boots the server.
3. Once the build finishes, you receive a secure **HTTPS URL**. Copy this URL.

> \[!NOTE]
> Ensure your deployment has the necessary environment variables set (like `API_KEY` or `DB_PASSWORD`) if your target API requires authentication. The MCP server will automatically proxy these credentials.

## Phase 3: Connecting the Agent

Now you must configure your AI client to communicate with your Karada endpoint.

### Claude Desktop Configuration

1. Open your Claude Desktop config file:
   * **Mac:** `~/Library/Application Support/Claude/claude_desktop_config.json`
   * **Windows:** `%APPDATA%\Claude\claude_desktop_config.json`
2. Add your Karada endpoint using the built-in MCP Server block:

```json theme={null}
{
  "mcpServers": {
    "my-company-api": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/client-cli", "connect", "https://your-karada-url.karada.ai/sse"],
      "env": {}
    }
  }
}
```

3. Save the file and restart Claude Desktop.
4. You should see a **Tools (🛠️)** icon in the input bar. Click it to verify your API endpoints are listed as available tools.

## Phase 4: Interacting with the Agent

Your agent is now fully connected. Try prompting it with tasks that require your API's capabilities.

* **Example Prompt**: "Look up the user '[alice@example.com](mailto:alice@example.com)' and tell me their subscription status."
* **Example Prompt**: "Fetch the latest 5 support tickets and summarize their issues."

The agent will autonomously decide which tool to call, format the payload, send the request to your Karada MCP Server, and process the response!