The standout is the Context MCP: you point Dial at your own MCP server and the voice agent calls your tools live, mid-call. That turns a phone call from a one-way notification into a two-way command channel into your own systems, I haven't seen another telephony API do this.
We built an AI on-call commander on it: it places a real outbound call to an engineer, and over that same call they can dismiss an alert, trigger a second call to a teammate, run a gated runbook, or escalate to a manager - all executed against production in real time, by voice.
Getting a real phone number for your agent is the other surprise, it takes seconds, not the usual telephony onboarding slog. Provision a number, drop your system prompt into the call instruction, and you've got a live agent on a real line. The UI/UX is smooth throughout, but the API documentation and SDK are the real standouts: clear, well-organized docs and a clean CLI meant we went from zero to real outbound calls fast, with no telephony plumbing. It's genuinely agent-native, works out of the box with Claude, Cursor, LangChain and MCP.
The support and Discord community were fantastic - responsive, hands-on, and quick to unblock us whenever we had a question. It's the kind of support that makes you confident building on a platform.
The ROI was positive almost immediately: the low barrier to start plus the fast integration meant we delivered a working, high-value product in a fraction of the time and cost we'd budgeted, and the real-time conversation quality held up to an actual high-stakes incident run entirely by voice. That combination of speed, low cost, capability, and support is rare in infrastructure. Review collected by and hosted on G2.com.
The one thing I'd tune: the voice agent will faithfully say whatever you give it, so out of the box a long system prompt can produce an opening that runs longer than you'd want on a live call. We got crisp results once we tightened our prompts and kept openings short, so it was really a prompting adjustment on our side, not a product flaw. A few built-in best-practice defaults or examples for voice phrasing would make that even smoother for first-timers. Review collected by and hosted on G2.com.