Skip to main content
Model Context Protocol (MCP) is an open standard that lets AI agents connect to external tools and data sources and execute actions against them in real time. In VoiceInfra, MCP is the bridge between your AI agents and the rest of your tech stack — not after a call, but during it. While the caller is still on the line, your agent can book an appointment in Google Calendar, update a Salesforce opportunity, check live inventory, or send a Slack notification. No “I’ll follow up with you” — the action happens now.

​
What You Can Do With MCP

  • Book appointments in Google Calendar or Calendly while the caller waits
  • Update Salesforce or HubSpot records mid-call without post-call data entry
  • Send Slack or Microsoft Teams notifications to your team in real time
  • Query live inventory levels, product pricing, or warehouse availability
  • Trigger workflows in Asana, Monday.com, Pipedream, or any MCP-compatible system
  • Look up customer account data from your internal database
  • Geocode addresses, check delivery status, or verify appointment availability
Compatible tools include (but are not limited to): Microsoft Teams, Asana, Pipedream, Salesforce, HubSpot, Slack, Zoom, and any system that exposes an MCP-compatible endpoint.

​
How MCP Works in Voice Calls

During a live call, the following happens in under a second:
  1. The caller says something that requires real-world action (e.g., “Can you book me in for Thursday at 3pm?”)
  2. The LLM decides to invoke the relevant MCP tool (e.g., book_appointment)
  3. VoiceInfra calls your MCP server with the required parameters
  4. Your MCP server executes the action and returns a result
  5. The LLM incorporates the result into its next spoken response (“Done — I’ve booked you in for Thursday at 3pm. You’ll receive a confirmation email shortly.”)
The caller experiences a natural, uninterrupted conversation. The action is complete before the agent finishes speaking.

​
Adding an MCP Server

1

Go to Settings → MCP Servers → Add MCP Server

Open your VoiceInfra dashboard and navigate to Settings → MCP Servers. Click Add MCP Server.
2

Enter server details

Fill in the following fields:
  • Server Name — a display name you’ll recognize (e.g., “Pipedream Workflows”)
  • Server URL — the HTTPS endpoint of your MCP server
  • Timeout — how long to wait for a response before giving up (default: 5 seconds, maximum: 600 seconds)
  • Description (optional) — notes about what this server provides
3

Add authentication headers

If your MCP server requires authentication, add the necessary HTTP headers (for example, Authorization: Bearer YOUR_TOKEN). Headers are encrypted before storage and are never exposed in logs or the dashboard UI.
4

Save the server

Click Save. The server is immediately available to assign to any AI agent in your workspace.
5

Assign the server to your agent

Open your agent’s configuration, go to the MCP tab, and select the server you just added. Save your agent configuration. The agent can now invoke tools from this server during live calls.

​
Configuration Reference

FieldDescription
Server NameDisplay name shown in the dashboard and agent configuration
Server URLThe HTTPS endpoint VoiceInfra calls when invoking a tool
TimeoutMaximum wait time for a server response (1–600 seconds; default 5s)
Auth HeadersCustom HTTP headers (e.g., Authorization, X-API-Key) sent with every request — stored encrypted
DescriptionOptional notes visible only to workspace admins
Set a conservative timeout (5–15 seconds) for most APIs to avoid pausing the conversation noticeably. Use a higher timeout only for slow internal systems, and consider setting those tools to async mode.

​
Pre-Built MCP Integrations

You don’t need to build your own MCP server to get started. Many popular platforms already offer hosted MCP servers you can connect with just a URL and an API key:
  • Pipedream — offers a hosted MCP server that gives your agents access to 2,800+ connected apps. Connect Pipedream workflows as callable tools with zero infrastructure to manage.
  • Zapier — supports MCP-compatible connections for agent tool use
  • n8n — can expose workflows as MCP tools from a self-hosted instance
For custom internal APIs, build a lightweight MCP server wrapper in front of your existing endpoints. Any language or framework works — all VoiceInfra requires is an HTTPS endpoint.

​
Security

  • Headers encrypted at rest — authentication headers are encrypted before being stored; they are only sent to your specified MCP server URL
  • Encrypted in transit — all API communications between VoiceInfra and your MCP server use HTTPS/TLS
  • No header exposure — headers are never shown in delivery logs, error messages, or the dashboard after saving
  • Admin-only configuration — only workspace owners and admins can add, edit, or delete MCP servers

​
Frequently Asked Questions

No. Many tools — including Pipedream, Zapier, and others — offer hosted MCP servers you can connect immediately. If you have an existing REST API, you can also wrap it with a thin MCP adapter rather than building from scratch.
There is no strict limit. Add as many servers as your use cases require. Manage all of them from the MCP Servers dashboard under Settings.
VoiceInfra handles failures gracefully. The agent waits for the configured timeout, then responds appropriately — informing the caller, retrying the request, or escalating to a human agent depending on your configuration. The call does not drop.
Yes. MCP is provider-agnostic. It works with all five LLM providers supported by VoiceInfra: OpenAI, Anthropic, Google Gemini, Groq, and Mistral.
Most well-optimized APIs respond in under 500ms, which is imperceptible to the caller. Configure a higher timeout for slower internal systems. For fire-and-forget operations that don’t need to return a result, consider using an async custom function instead.
Yes. Edit the MCP server configuration at any time from the MCP Servers dashboard. New headers take effect immediately for all subsequent requests.