Your AI assistant can build agents too
The voice agent MCP server gives Claude Code, Claude Desktop, and Cursor 14 tools to read the docs and build or edit agents that render on the visual canvas.
Developers
Everything your team does in the PeachDesk dashboard, your code can do through a fully documented REST API, typed Python and TypeScript SDKs, signed webhooks, an embeddable voice widget, and an MCP server for AI coding assistants. Own the stack. Do not just rent the pipeline.
PeachDesk is API-first. A fully documented REST API, versioned under/api/v1, covers agents, calls, campaigns, telephony configuration, knowledge bases, tools, credentials, organisations, usage, and billing. If the dashboard can do it, there is an endpoint for it.
Create API keys in the dashboard under Settings. Each key is scoped to your organisation, shown once at creation, and sent in theX-API-Key header. A default key is minted at signup, so your first request is minutes away.
Browse the API reference for endpoint-level documentation that stays mechanically in sync with the code.
curl https://api.peachdeskai.com/api/v1/agents \
-H "X-API-Key: $PEACHDESK_API_KEY"The Python and TypeScript SDKs are more than HTTP clients. Each one is a typed workflow builder that fetches the platform node specification and validates every call at the call site, in your editor or CI, instead of at runtime. Validation errors carry hints written for both humans and AI assistants to self-correct.
Both SDKs speak the same agent format as the dashboard, so an agent built in code opens on the visual canvas and an agent built on the canvas round-trips through code. Access details and worked examples are in the SDK documentation.

Subscribe to call_started, call_ended, andcall_analyzed events per agent. Every delivery is signed with HMAC-SHA256 so your receiver can verify it came from PeachDesk, and a test endpoint sends a signed payload so you can validate your handler before the first real call.
Need to call out instead? A webhook node inside any workflow sends a templated HTTP request when the conversation reaches it, authenticated with credentials from an encrypted vault. The two mechanisms are independent: events notify you, nodes notify your systems.
The embeddable widget puts a live voice agent on your site in three modes: a floating button, an inline panel inside your own container, or a headless mode driven entirely from JavaScript. You restrict the widget to approved domains. It is voice only, by design: it starts and ends real calls with your published agent.
<script
src="https://app.peachdeskai.com/embed/peachdesk-widget.js?token=YOUR_EMBED_TOKEN"
async
></script>The voice agent MCP server gives Claude Code, Claude Desktop, and Cursor 14 tools to read the docs and build or edit agents that render on the visual canvas.
With voice AI BYOK, each pipeline stage runs on your own provider accounts. BYOK calls never consume PeachDesk credits.
The developer surface sits on top of the full voice AI platform: a visual agent builder, telephony, campaigns, and two model planes, one platform.
Create an organisation-scoped API key in the dashboard under Settings. The key is shown once, so store it somewhere safe. Send it on every request in the X-API-Key header. A default key is created when you sign up, and you can create, list, archive, and reactivate keys as your integration grows.
Yes. PeachDesk provides typed SDKs for Python and TypeScript that double as workflow builders. Both validate every agent definition against the platform node specification at the call site, so mistakes surface in your editor or CI instead of at runtime. The documentation covers access and a worked example for each language.
Yes. Add one script tag and the widget renders a floating voice button, an inline panel inside your own container, or a headless mode you drive from JavaScript. You restrict the widget to approved domains. The widget is voice only: it starts and ends live calls with your published agent.
Yes. The PeachDesk MCP server gives assistants like Claude Code, Claude Desktop, and Cursor 14 tools to read the documentation and create or edit agents. The assistant emits typed SDK code, the platform parses and validates it without executing it, and the result appears on the same visual canvas your team uses.
Read the documentation, mint a key, and make your first call today.