MCP integration
Connect to your hosted MCP Server using any MCP client to communicate with tools. Use MCP from Zoft voice agents, chat agents, workflows and AI employees.
- Starts withMCP Tool
- ZoftAgent or workflow decides what happens next
- Then in MCPReply to MCP Client
Zoft + MCP
What Zoft can do with MCP.
- 01Voice agentsA phone agent can use MCP actions such as Reply to MCP Client during the call, then hand off to a person when it should.
- 02Chat agentsThe same agent on web chat, WhatsApp or SMS can look up and update MCP while it talks to the customer.
- 03WorkflowsStart a workflow when “MCP Tool” happens in MCP, or use MCP as any step on the canvas.
- 04AI employeesAn AI employee can find and use MCP actions on its own, and ask for your approval before it sends or changes anything important.
Actions 1
Things Zoft can do in MCP.
- Reply to MCP ClientReturn a response to the MCP client that called the tool.
Triggers 1
Events in MCP that can start a workflow.
- MCP ToolCreates a tool that MCP clients can call to execute this flow
Setup
Connect MCP in three steps.
- 01ConnectMCP is built in. There is nothing to connect, so you can use it in a workflow straight away.
- 02Put it to workAdd MCP as a workflow step, or let a voice agent, chat agent or AI employee use it as a tool.
- 03Review every runAdd approvals where you want a person in the loop. Every run leaves a record you can check.
More ai and models integrations.
See all in AI and models
FAQ
MCP on Zoft, answered.
What can Zoft do with MCP?
Zoft supports 1 action and 1 trigger for MCP. You can use them as workflow steps, and voice agents, chat agents and AI employees can call them as tools.
How do I connect MCP to Zoft?
MCP is built in. There is nothing to connect, so you can use it in a workflow straight away.
Can a voice agent or chat agent use MCP?
Yes. Once MCP is connected, a voice agent or chat agent can use its actions during a conversation. Permissions control which actions each agent can use.
Can an AI employee work in MCP without supervision?
Only as far as you allow. AI employees ask for approval before sensitive steps, and every run leaves a record you can review.