Voice agent MCP
Voice agent MCP: your AI assistant can build your voice agents
The PeachDesk MCP server connects Claude Code, Claude Desktop, and Cursor to the platform. Your assistant reads the documentation, writes typed SDK code, and saves a validated agent that renders on the same visual canvas your team edits by hand.
Fourteen tools, one specification
The server exposes 14 tools over streamable HTTP: documentation search and reading, agent creation and editing, and the platform catalogs for node types, tools, documents, credentials, and recordings. Authentication reuses the organisation API key from the developer platform, so there is nothing new to manage.
Because the tools validate against the same node specification that drives the dashboard and the SDKs, an agent your assistant drafts is an agent the platform can actually run.
{
"mcpServers": {
"peachdesk": {
"url": "https://api.peachdeskai.com/api/v1/mcp",
"headers": {
"X-API-Key": "paste-your-api-key-here"
}
}
}
}Full setup steps for each assistant live in the MCP server documentation.
Safe by construction
The assistant writes typed TypeScript SDK code, but PeachDesk never executes it. The platform parses the code, validates every node and connection against the platform specification, and only then saves the agent. Validation errors carry hints the assistant uses to correct itself on the next attempt.
The result is not a black box. Every assistant-built agent opens on the visual canvas, where your team reviews, edits, tests, and publishes it like any other agent.
Works where you already work
Claude Code for terminal workflows, Claude Desktop for conversational drafting, and Cursor for editor-native building. Point any of them at the same endpoint with the same API key. Ask for a support agent that answers from your knowledge base, and watch the draft appear on the canvas.
Want the concepts first? Read how a voice agent MCP server works, then come back to connect your assistant.
MCP questions, answered
What is the PeachDesk MCP server?
It is a Model Context Protocol server built into the PeachDesk platform. MCP is a standard way to give AI assistants tools. The PeachDesk server offers 14 tools that let an assistant search the documentation, inspect the node catalog, and create or edit voice agents, all authenticated with your organisation API key.
Which AI assistants work with the PeachDesk MCP server?
Claude Code, Claude Desktop, and Cursor are the documented setups. Any assistant that can reach an MCP server over HTTP with a custom header can use the same endpoint, because authentication is your standard PeachDesk API key sent as a header on each request.
Does the MCP server execute code the assistant writes?
No. The assistant writes typed TypeScript SDK code, and the platform parses it, validates every node and connection against the platform specification, and only then saves the agent. Generated code is never executed on PeachDesk servers, and the finished agent opens on the visual canvas for your review.
How do I connect Claude Code or Cursor to PeachDesk?
Add the PeachDesk MCP endpoint to your assistant configuration with your API key as a header. The endpoint URL and a configuration block you can copy for each supported assistant are in the documentation. After that, describe the agent you want and your assistant drafts it for you.
Hand your assistant the keys to the canvas
Connect the MCP server in minutes, then describe your first agent.