Recepta.ai exposes a Model Context Protocol server, so an AI assistant can query your workspace in natural language — “how many calls did we miss last week”, “show me the transcript where the customer mentioned a leak” — without you building an integration.
Endpoint
Streamable HTTP transport, stateless. Authenticate with an API key in the x-api-key header. The key’s permissions govern what the assistant can reach — a key without calls:transcript means no transcript access, regardless of what’s asked for.
Connecting a client
Most MCP clients take a URL and headers. In Claude Code:
For other clients, consult their MCP configuration — you’ll supply the URL above and the x-api-key header.
Add x-location-id as a second header to scope the whole session to one location. Calls
SMS
Campaigns
Agents & analytics
Read-only by design
These tools query your workspace; they don’t place calls, send messages, or modify records. An assistant connected this way can analyze what happened but can’t act on your customers.
Use a dedicated API key for MCP with read scopes only — calls:read, contacts:read, sms:read, analytics:read. Add calls:transcript only if you want the assistant reading what customers said.
Worth asking it
- “What were the most common reasons callers were transferred to a human last month?”
- “Which calls ended without a booking, and what did the customer ask about?”
- “Compare booking rates across our locations this quarter.”
- “Find SMS conversations where someone asked about pricing but never booked.”
These are the questions that normally require exporting data and building a report. The answers point straight at what to fix — see QA & learning.
Call transcripts contain customer personal data. Connecting an assistant with transcript access means that data goes to whichever AI provider powers it. Confirm that’s compatible with your privacy policy and customer commitments. See Security & privacy. Rate limits
MCP requests are rate limited per API key by plan tier, as the REST API is. See Rate limits. 