PolyAI lets developers build voice agents by talking to their coding assistant

What PolyAI released

PolyAI, a company that builds AI agents that talk with customers, has launched the Builder MCP. As poly.ai announced, it lets developers create, configure, test and deploy what it calls dialog agents (AI built for back-and-forth conversation, such as on a phone line) using natural language, from inside the AI coding assistant they already use.

The company describes the problem it is fixing as a kind of tax. Until now, building one of these agents meant working in two places at once: logic in a code editor, and configuration in a web dashboard called Dialog Studio. The alternative was writing raw calls against PolyAI's API (the technical doorway software uses to talk to a service) by hand and checking every response yourself.

What MCP means and how it works

MCP stands for Model Context Protocol, an open standard introduced by Anthropic for connecting AI agents to outside tools and data. Think of it as a shared plug. A server lists the tools it offers, the AI agent reads that list, and it picks the right tool when you make a plain-language request.

PolyAI's server wraps its entire REST API (the standard web-based interface to the platform) in that plug. So every action the API allows becomes something your AI assistant can do for you. The company stresses that nothing changes under the hood; it is a new way into the same platform.

What the agent can do for you

The list of abilities is broad. According to the announcement, an assistant connected to the Builder MCP can:

• create, configure, duplicate and delete agents
• manage knowledge-base topics and rules for agent behavior
• version agents with branches, including create, merge and sync
• manage variants, config pages and settings for each environment
• import and manage phone numbers and telephony connectors
• publish, promote and roll back deployments across environments
• pull conversations and audio recordings, and set alert rules
• place outbound calls once telephony is set up for the agent

PolyAI argues this breadth is the point. It says many voice platforms offer an MCP for the basics, while this one covers what enterprise teams need to go to production, such as versioning, rollback and monitoring. Because it wraps the API instead of copying a slice of it, the company says new platform features will reach the MCP without extra integration work.

A walkthrough and how to start

PolyAI's example hands Claude Code one set of instructions: list the agents in an empty account, create a dental-office agent that greets callers by the practice's name, test it in a chat session, version it on a branch, add knowledge, then delete it. The company says that covers the full lifecycle without opening the dashboard or writing an HTTP request. A video walkthrough is on Loom. The post itself notes that its sample responses are representative and its IDs illustrative.

The server is available in all regions and connects with a single command. You sign in with a workspace-scoped API key, created from the API Keys tab in Dialog Studio. You choose which agents it can reach and what it may do, and the full key is displayed only once.

PolyAI advises giving the key the least access it needs and storing it in your client's secret settings, not pasting it into a prompt. The key's region must match your workspace. Besides Claude Code, Cursor, Claude Desktop and Codex are supported with the same URL and headers.

Who it is for and what to watch

This is a developer tool. If you never build agents, the newsletter's advice stands: you can skip it. For teams that do, the appeal is fewer tool switches and a quicker loop between describing an agent and testing it.

The security detail deserves attention. An AI assistant that can delete agents, deploy changes and place outbound calls is powerful, so narrow permissions matter. PolyAI says OAuth, a sign-in method that avoids pasting long-lived keys, is coming, but for now API keys are the self-serve option. Pricing is not stated, and the company asks users to send feedback to shape what comes next.