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

# MCP Server

> Manage UponAI voice agents directly from AI coding tools like Cursor, Claude Desktop, and Claude Code using the Model Context Protocol.

UponAI supports the Model Context Protocol (MCP), letting you build and manage voice agents directly from MCP-capable clients — Cursor, Claude Desktop, Claude Code, and similar tools.

If you already use UponAI via the REST API or SDKs, MCP is simply a different interface to the same functionality, optimized for agentic workflows in IDEs and AI assistants.

## What the MCP Server Can Do

With an MCP client connected, your AI assistant has native access to:

| Category              | Actions                                                                      |
| --------------------- | ---------------------------------------------------------------------------- |
| **Agents**            | Create, update, publish, list, fetch versions                                |
| **Calls**             | Create phone/web calls, fetch details, list, delete, update metadata, run QA |
| **Phone numbers**     | Import, provision, list, fetch                                               |
| **Knowledge base**    | Create KBs, attach/remove sources, list                                      |
| **Voices**            | List, clone, search community voices                                         |
| **Chats**             | Create chats and chat agents, end chats, update metadata                     |
| **Testing & QA**      | Create test cases, run tests, list/rerun QA, submit scores                   |
| **Alerts & webhooks** | Create/list alert rules, list incidents, test webhooks                       |

## Prerequisites

* A UponAI API key (from the **API Keys** tab in your dashboard)
* An MCP client (Cursor, Claude Desktop, Claude Code, etc.)
* MCP server URL: `https://retell.stlmcp.com`
* Auth header: `Authorization: Bearer <YOUR_API_KEY>`

## Setup

<Tabs>
  <Tab title="Cursor">
    Open the command palette → **Cursor Settings → MCP → Add new global MCP server**:

    ```json theme={null}
    {
      "mcpServers": {
        "uponai": {
          "url": "https://retell.stlmcp.com",
          "headers": {
            "Authorization": "Bearer <YOUR_API_KEY>"
          }
        }
      }
    }
    ```
  </Tab>

  <Tab title="Claude Desktop">
    Add to your Claude Desktop MCP config:

    ```json theme={null}
    {
      "mcpServers": {
        "uponai": {
          "url": "https://retell.stlmcp.com",
          "headers": {
            "Authorization": "Bearer <YOUR_API_KEY>"
          }
        }
      }
    }
    ```
  </Tab>

  <Tab title="Claude Code">
    Run in your terminal:

    ```bash theme={null}
    claude mcp add --transport http uponai https://retell.stlmcp.com \
      --header "Authorization: Bearer <YOUR_API_KEY>"
    ```
  </Tab>

  <Tab title="Other clients">
    Configure with:

    * **Transport:** Streamable HTTP
    * **URL:** `https://retell.stlmcp.com`
    * **Header:** `Authorization: Bearer <YOUR_API_KEY>`
  </Tab>
</Tabs>

## Example Prompts

Once connected, try these prompts in your MCP client:

* *"List my agents and summarize what each one does."*
* *"Create a new agent for inbound sales qualification and publish it."*
* *"Show me the last 20 calls and flag any with low QA scores."*
* *"Create a knowledge base and attach these sources, then update my agent to use it."*
* *"Import this phone number and assign it to my agent."*
* *"Rerun QA on this call and summarize the failure reasons."*

## Security

Connecting an LLM to operational tools introduces prompt injection risk — untrusted content in call transcripts, user messages, or KB documents can include instructions that try to trick the model into taking unintended actions.

<Warning>
  Keep "confirm before running tools" enabled in your MCP client and review tool calls carefully before approving.
</Warning>

**Recommendations:**

* Use least privilege — create API keys with minimum required permissions
* Never paste API keys into chat — store them in client secrets/settings
* Prefer read-first workflows — fetch a resource before updating or deleting it
* Gate destructive actions — delete and publish should require explicit intent
* Avoid sending full call transcripts to the model if you don't need them — they may contain PII
* Use development data when exploring agent behavior, not production

## Troubleshooting

| Issue                | Solution                                                                       |
| -------------------- | ------------------------------------------------------------------------------ |
| Tools not showing up | Verify the server URL, auth header, and that the client can reach the endpoint |
| 401 Unauthorized     | Confirm `Authorization: Bearer <YOUR_API_KEY>` and that the key is active      |
| Tool call errors     | Re-run with smaller inputs and inspect the returned error payload              |