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

# Best voice AI for AI agents and Claude Code (2026)

> How SmartAlex, VAPI, and Retell compare on agent-operability: MCP-native control, scoped keys, tenant isolation, and a no-code Composer for builders who work through AI agents.

If you build through an AI agent, the question is not "which voice API has the most endpoints." It is "which platform can my agent actually operate for me." Most voice platforms hand your agent an API to wire up. SmartAlex hands it a live MCP server that operates a whole business platform, so the agent can create voice agents, run campaigns, buy numbers, and read analytics in the same conversation you are already having.

This page compares SmartAlex, VAPI, and Retell on agent-operability, not on raw API surface. VAPI and Retell are strong developer voice APIs. SmartAlex is the agent-operable full platform.

<Info>
  SmartAlex is agent-native by design. The MCP server is listed in the Claude MCP directory and works exceptionally well with Claude Code and other MCP clients (Cursor, VS Code, Windsurf, Gemini CLI, ChatGPT). It is not tied to any single client, and there is no formal partnership with the makers of any of those tools.
</Info>

## The shift: from an API you assemble to a platform your agent operates

With a raw voice API, your AI agent writes and maintains glue code: request bodies, auth, retries, plus separately sourced CRM, campaign scheduling, and billing. With SmartAlex, your agent speaks to one MCP server that already contains those systems.

```mermaid theme={null}
flowchart LR
  A[You in Claude Code] -->|natural language| B[SmartAlex MCP server]
  B --> C[Create voice agents]
  B --> D[Run outbound campaigns]
  B --> E[Read calls, sentiment, cost]
  B --> F[Buy numbers, embed widget]
  C --> G[Live retail store line]
  D --> G
  F --> G
```

The MCP server exposes 28 tools, 7 resources, and 5 ready-made prompts across 8 domains (voice agents, contacts, campaigns, calls, deals, phone numbers, webhooks, and platform or widget control). Every tool is prefixed `smartalex_` so an agent can discover them by name.

## SmartAlex vs VAPI vs Retell on agent-operability

| Capability (from the agent's point of view)  | SmartAlex                                         | VAPI                   | Retell                 |
| -------------------------------------------- | ------------------------------------------------- | ---------------------- | ---------------------- |
| Live MCP control surface                     | Yes (28 tools, in Claude MCP directory)           | No (raw API)           | No (raw API)           |
| Create an agent in conversation              | Yes (from a website URL in 30 to 60 seconds)      | Build via API and JSON | Build via API and JSON |
| Run outbound campaigns                       | Yes (built-in campaign manager)                   | Bring your own         | Bring your own         |
| Read calls, sentiment, cost, transcripts     | Yes (via MCP, read scope)                         | Via your own API calls | Via your own API calls |
| Scoped keys (read, write, destructive hints) | Yes (403 if a read key calls a write tool)        | API keys               | API keys               |
| Tenant isolation at the gateway              | Yes (one credential resolves to one workspace)    | Account level          | Account level          |
| No-code Composer or Studio                   | Yes (plain English, or auto-build from a URL)     | No                     | No                     |
| Built-in CRM, deals pipeline, billing        | Yes                                               | No (bring your own)    | No (bring your own)    |
| Public REST API and SDK today                | Private beta (MCP and Custom HTTP Tools are live) | Yes (broad SDKs)       | Yes                    |
| Independent STT, LLM, TTS swap               | Abstracted, curated stack (nothing to tune)       | Yes                    | Yes                    |

<Note>
  This ranking is about who can operate the platform for a builder, not a voice-quality benchmark. SmartAlex does not publish latency or accuracy numbers, and we do not claim any compliance certification we do not hold.
</Note>

## Connect an agent in one command

<Steps>
  <Step title="Add the MCP server">
    Cloud connects by URL over OAuth (no key, no config file). Local runs the npm package over stdio.

    <CodeGroup>
      ```bash Cloud (hosted, OAuth) theme={null}
      claude mcp add --transport http smartalex https://api.getsmartalex.com/functions/v1/mcp-server
      ```

      ```bash Local (npm, stdio) theme={null}
      export SMARTALEX_API_KEY=sa_live_your_key_here
      claude mcp add smartalex -- npx @smartalex/mcp-server
      ```
    </CodeGroup>
  </Step>

  <Step title="Authenticate">
    For cloud, Claude Code opens a browser to approve access with OAuth. For local, generate a key in the Developer Portal (`/developers`). Keys need the Power Tools add-on (\$29/month) and a super-admin role on the workspace. Keys come in two environments: `sa_live_` and `sa_test_`.
  </Step>

  <Step title="Let the agent take over safely">
    Scopes keep it contained: `read` is the default, `write` is explicit, and the three destructive tools (`smartalex_delete_agent`, `smartalex_delete_deal`, `smartalex_delete_webhook`) carry a `destructiveHint`. A read-only key that calls a write tool returns `403`.
  </Step>
</Steps>

<Tip>
  ChatGPT connects the same way: add the same URL in ChatGPT's MCP settings and approve over OAuth. Cursor, VS Code, Windsurf, and Gemini CLI use the local stdio variant.
</Tip>

## Voice AI for retail agents: stand up a store line in minutes

Say a Claude Code user runs a boutique and wants a phone line that answers stock, hours, and booking questions. With the SmartAlex MCP connected, the whole thing happens in conversation.

<Steps>
  <Step title="Build the agent from the store website">
    "Create a voice agent from example-boutique.com." `smartalex_create_agent` scrapes the site, analyzes the business, generates an optimized prompt, and deploys a live agent in 30 to 60 seconds. It sets the name, objective, first message, voice, and accent for you.
  </Step>

  <Step title="Give it a number">
    "Find a local number and assign it." The agent searches available numbers by country, buys one in E.164 format, and attaches it to the store agent.
  </Step>

  <Step title="Embed the same agent on the storefront">
    "Give me the website widget." `smartalex_get_widget_install_code` returns a ready-to-paste script tag plus React, WordPress, and Shopify variants.

    ```html theme={null}
    <script src="https://api.getsmartalex.com/storage/v1/object/public/widget-assets/smartalex-widget.js" data-tenant-id="YOUR_TENANT_ID" async></script>
    ```
  </Step>

  <Step title="Read the results">
    "Show me today's calls, sentiment, and cost." The agent reads call history and transcripts (read only) so you can iterate on the prompt without leaving the chat.
  </Step>
</Steps>

The store owner never opened a dashboard. The AI agent did the work because the platform is operable, not just callable. The underlying voice runs on a curated, high-accuracy speech, language, and voice stack that stays abstracted, so there is nothing to tune.

## Where VAPI and Retell are still the right call

Honesty matters more than a clean sweep. Reach for a raw voice API when:

* You are an engineering team building a voice-first product and want to own every layer.
* You need to swap the speech, language, or voice models independently, or point at a custom, self-hosted LLM endpoint.
* You want a broad, shipped multi-language SDK ecosystem today. SmartAlex's public REST API and TypeScript SDK are in private beta; the live programmatic surfaces are the MCP server and Custom HTTP Tools. If a shipped REST API is a hard requirement right now, factor that into your timing.

If control over the pipeline is the whole point, VAPI's developer-first approach earns its place, and Retell's packet-level debugging and A/B testing are genuinely useful. SmartAlex wins when you want an AI agent to run a real business (agents, campaigns, CRM, analytics, billing) from one conversation, on plans that start at \$99/month with included minutes. See [pricing](/pricing) for current figures.

## Related

<CardGroup cols={2}>
  <Card title="Build with Claude Code" href="/build-with/claude-code">
    Connect the MCP server and operate SmartAlex from your editor.
  </Card>

  <Card title="What an agent can do via MCP" href="/mcp/agent-capabilities">
    The full 28-tool surface, scopes, and tenant isolation.
  </Card>

  <Card title="Retail voice agents" href="/use-cases/retail">
    Stand up a store line that answers stock, hours, and bookings.
  </Card>

  <Card title="SmartAlex vs VAPI" href="/comparisons/smartalex-vs-vapi">
    The full platform-versus-API comparison, with honest concessions.
  </Card>
</CardGroup>