SAI VIGNESH M
Updated on 30-07-2026
The purpose of this page is to clarify a common point of confusion when working with Knit and June AI. What Knit refers to as an MCP Server is referred to as a Connector within June AI. This difference in naming can be confusing, especially for users who are new to the June AI platform.
In simple terms, June AI consumes MCP services provided by Knit. The connection is created from within an individual agent, rather than from a global or system-level settings page.
Once the connection is established, the June AI Agent can use the tools that are hosted on the Knit MCP Server. There is no additional middleware involved. All interactions occur through direct calls made by the agent to the permitted MCP endpoints.
If functionality issues occur after setup, they are rarely caused by MCP itself. In most cases, the problem is related to missing permissions or an incorrect access token.
These checks are basic, but verifying them early will prevent unnecessary troubleshooting later.
MCP settings are not located in general system or application settings. Instead, MCP is attached directly to each agent that you create.
To add an MCP Server, open the agent you want to configure and navigate to the Connectors section. From there, select the option to add a new connector.
Follow the steps below to add a Knit MCP Server as a connector:
If the connector fails to save at this stage, stop and resolve the issue immediately. Continuing with the setup without a successful save will not resolve the problem.
After the MCP Connector has been created, not all tools are available to the agent by default. Each tool must be explicitly enabled or granted permission before it can be used.
If you encounter a "Tool Not Found" error, this usually indicates that the connector itself is working correctly, but permission for the specific tool has not yet been enabled.
When testing the connection, keep the initial test simple. Make a single request and verify that a single response is returned successfully.
Once the basic test succeeds, you can proceed with more complex calls and workflows using the MCP Connector.
MCP Connectors are generally very stable in June AI once configured correctly. Most issues arise from rushing through the initial setup or skipping the permissions step. Taking a few extra minutes to verify configuration and permissions helps avoid repeated issues later.
