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

# MCP Server

> Connect AI coding tools to Endprompt via Model Context Protocol

Endprompt exposes its Admin API as an [MCP (Model Context Protocol)](https://modelcontextprotocol.io/) server, allowing AI coding tools to manage your endpoints, prompts, and executions directly through natural conversation.

## What is MCP?

MCP is an open protocol that lets AI assistants use external tools. When you connect Endprompt's MCP server to your coding tool, the AI can:

* Create and configure endpoints
* Write and test Liquid prompt templates
* Promote prompts to Live and set defaults
* Execute endpoints and analyze results
* Monitor stats and browse execution logs

All without you leaving your editor.

## Supported Tools

The MCP server works with any client that supports the Streamable HTTP transport:

* **VS Code** with GitHub Copilot or compatible extensions
* **Claude Desktop**
* **Cursor**
* **Windsurf**
* Any MCP-compatible client

## Setup

### Prerequisites

* An [Admin API key](/admin-api/authentication) with the `epa_` prefix
* An MCP-compatible client

### VS Code

Create a `.vscode/mcp.json` file in your project:

```json theme={null}
{
  "servers": {
    "endprompt": {
      "type": "http",
      "url": "https://api.endprompt.app/mcp",
      "headers": {
        "Authorization": "Bearer ${input:endpromptApiKey}"
      }
    }
  },
  "inputs": [
    {
      "id": "endpromptApiKey",
      "type": "promptString",
      "description": "Endprompt admin API key (epa_...)",
      "password": true
    }
  ]
}
```

VS Code will prompt you for the API key when the MCP server is first used. The key is stored securely per session.

<Tip>
  Add `.vscode/mcp.json` to your `.gitignore` to avoid committing it, or use the `${input:...}` pattern shown above to avoid hardcoding the key.
</Tip>

### Claude Desktop

Add to your `claude_desktop_config.json`:

<CodeGroup>
  ```json macOS theme={null}
  // ~/Library/Application Support/Claude/claude_desktop_config.json
  {
    "mcpServers": {
      "endprompt": {
        "url": "https://api.endprompt.app/mcp",
        "headers": {
          "Authorization": "Bearer epa_your_admin_key_here"
        }
      }
    }
  }
  ```

  ```json Windows theme={null}
  // %APPDATA%\Claude\claude_desktop_config.json
  {
    "mcpServers": {
      "endprompt": {
        "url": "https://api.endprompt.app/mcp",
        "headers": {
          "Authorization": "Bearer epa_your_admin_key_here"
        }
      }
    }
  }
  ```
</CodeGroup>

### Cursor

Add an MCP server in Cursor's settings with:

* **URL**: `https://api.endprompt.app/mcp`
* **Header**: `Authorization: Bearer epa_your_admin_key_here`

## Available Tools

Once connected, your AI assistant has access to 25 tools organized into three categories.

### Endpoint Management

| Tool               | Description                                              |
| ------------------ | -------------------------------------------------------- |
| `list_endpoints`   | List all API endpoints for your tenant                   |
| `get_endpoint`     | Get full endpoint details including field schemas        |
| `create_endpoint`  | Create a new API endpoint                                |
| `update_endpoint`  | Update endpoint metadata (name, description, visibility) |
| `delete_endpoint`  | Soft-delete an endpoint                                  |
| `add_input_field`  | Add a typed input field to an endpoint                   |
| `add_output_field` | Add a typed output field to an endpoint                  |

### Prompt Lifecycle

| Tool                    | Description                                         |
| ----------------------- | --------------------------------------------------- |
| `list_prompts`          | List prompts for an endpoint with status filtering  |
| `get_prompt`            | Get full prompt details including Liquid template   |
| `create_prompt`         | Create a new prompt in Draft status                 |
| `update_prompt`         | Update template or config (auto-versions on change) |
| `promote_prompt`        | Promote from Draft to Live                          |
| `set_default_prompt`    | Set a Live prompt as the endpoint default           |
| `archive_prompt`        | Archive a prompt                                    |
| `unarchive_prompt`      | Restore an archived prompt to Draft                 |
| `duplicate_prompt`      | Clone a prompt as a new Draft                       |
| `list_prompt_versions`  | View version history of a prompt                    |
| `validate_template`     | Check Liquid syntax and variable references         |
| `list_available_models` | List all available LLM models                       |

### Execution & Analytics

| Tool                  | Description                                                |
| --------------------- | ---------------------------------------------------------- |
| `execute_endpoint`    | Execute an endpoint with inputs, get LLM output            |
| `get_endpoint_stats`  | Execution stats (requests, latency, tokens, costs)         |
| `get_prompt_stats`    | Compare prompt performance within an endpoint              |
| `list_execution_logs` | Browse execution history with filtering                    |
| `get_execution_log`   | Full execution details (inputs, rendered prompt, response) |

## Example Conversations

### "Create a sentiment analysis endpoint"

The AI assistant will:

1. Call `list_available_models` to show you model options
2. Call `create_endpoint` with your chosen name and path
3. Call `add_input_field` to define the text input
4. Call `add_output_field` for sentiment and confidence outputs
5. Call `create_prompt` with a Liquid template
6. Call `validate_template` to check for errors
7. Call `promote_prompt` to make it Live
8. Call `set_default_prompt` to activate it

### "How are my endpoints performing?"

The AI assistant will:

1. Call `get_tenant_stats` for an overview
2. Call `list_endpoints` to show your endpoints
3. Call `get_endpoint_stats` for specific endpoints you ask about
4. Call `get_prompt_stats` to compare prompt performance

### "Show me recent errors"

The AI assistant will:

1. Call `list_execution_logs` filtered to recent entries
2. Call `get_execution_log` for any failures to show full details
3. Analyze the rendered prompt and raw response to identify issues

## Transport Details

| Setting       | Value                               |
| ------------- | ----------------------------------- |
| **Protocol**  | MCP (Model Context Protocol)        |
| **Transport** | Streamable HTTP                     |
| **URL**       | `https://api.endprompt.app/mcp`     |
| **Auth**      | `Authorization: Bearer epa_...`     |
| **Session**   | Managed via `Mcp-Session-Id` header |

The server also supports legacy SSE transport for older MCP clients.