MCP Servers

The Model Context Protocol (MCP) allows your agents to connect to external tool servers, extending their capabilities beyond built-in tools. With MCP, your agents can access proprietary APIs, databases, internal systems, and third-party services, all through a standardised interface.

What is MCP?

MCP is an open protocol that standardises how AI models connect to external tools and data sources. Rather than hard-coding integrations, MCP allows your agent to dynamically discover and use tools provided by an MCP server at runtime. Each server exposes a set of tools with defined input schemas, and the agent can invoke them as part of a live conversation.

This means you can extend your agent's abilities without modifying the agent itself; simply point it at a new MCP server and the tools become available immediately. Common use cases include querying internal databases, triggering workflows in external systems, fetching real-time data from APIs, and interacting with custom business logic.

Advanced feature

MCP is an advanced feature designed for teams that need to integrate custom tooling or connect agents to proprietary systems. If you're looking for common integrations like calendar booking or CRM updates, check the built-in Tools and Integrations pages first.

Adding an MCP Server

Follow these steps to connect an MCP server to your agent:

Open the agent editor

Navigate to the Agents page and click on the agent you want to configure. This opens the agent editor where you can modify all aspects of the agent's configuration.

Go to the MCP tab

In the left sidebar of the agent editor, click the MCP tab. This is where all MCP server connections for this agent are managed.

Toggle MCP on

If MCP is not already enabled, toggle the MCP switch to the on position. This activates MCP support for the agent and reveals the server management interface.

Click Add Server

Click the Add Server button to open the server configuration form. You can add multiple servers; each one contributes its own set of tools to the agent.

Provide server details

Fill in the required fields:

Name: A descriptive label for this server connection (e.g. "Inventory API" or "Internal CRM").
URL: The endpoint URL of the MCP server.
Timeout (optional): How long to wait for a response before the request times out.
Session Timeout (optional): How long an MCP session persists before it is automatically closed.

Click Save to add the server. The agent will discover the server's available tools automatically.

Configuration

Each MCP server connection is defined by the following fields:

Field	Required	Description
Name	Yes	A descriptive label that identifies this server in the agent editor. Choose something meaningful, e.g. "Order Management" or "Weather API".
URL	Yes	The MCP server endpoint. This is the URL your agent will connect to when invoking tools. Must be a valid, reachable HTTP or HTTPS endpoint.
Timeout	No	The maximum time (in seconds) the agent will wait for a response from the MCP server before timing out. Useful for servers with variable latency.
Session Timeout	No	How long (in seconds) an MCP session remains active before it is automatically closed. Sessions persist across multiple tool invocations within a single call, so set this longer than your expected call duration.

Managing Servers

All MCP server connections for an agent are visible in the MCP tab of the agent editor. From here you can:

View: See the name, URL, and timeout configuration for each connected server at a glance.
Edit: Click on any server to update its name, URL, or timeout settings. Changes are saved to the agent configuration immediately.
Remove: Delete a server connection when it is no longer needed. The agent will lose access to that server's tools immediately.

MCP servers are versioned alongside your agent configuration. This means any changes you make (adding, editing, or removing servers) are tracked as part of the agent's overall configuration history, making it straightforward to audit and roll back if necessary.

Test before deploying

After adding or modifying an MCP server, use the browser-based Test Call to verify that the agent can discover and invoke the server's tools correctly. This ensures everything is working before real callers interact with the agent.