MCP Settings

The MCP server is a Beta feature for environments with UnqorkAI. Test each published tool and verify its results before relying on it.

MCP Settings is the page of a module's Settings dialog where Creators publish an API module as a tool for external AI agents. Creators also edit the tool's contract there. The page displays only for API modules in environments with UnqorkAI. For how the MCP server works, see MCP Server for API Modules. For the steps to publish a tool, see Publish API Modules as MCP Tools.

To open MCP Settings:

In the Explorer panel, select (ellipsis) next to an API module and select Settings.
In the left rail of the dialog, select MCP Settings.

The module must be open for editing to change these settings. The page saves together with the dialog's other pages when a Creator selects Save.

Tool Availability

The top of the page controls whether agents can see the tool. It also displays what an agent needs before it can call the tool.

Setting	Description
Before an agent can call this tool	An alert that lists what every agent needs: an Express role of its own with Default Permission No Access, Write on each module it calls, and an API credential that holds only that role. The page always displays this alert.
How to set up an agent	A collapsible guide to the steps from creating the role to connecting the agent. It also displays the environment's MCP endpoint and a copyable token request. The guide is open while Customize RBAC for This Module is off or no role can call the tool, and closed otherwise.
Agents can call this module as a tool (MCP)	When set to ON, this setting publishes the module as a tool. When set to OFF, agents no longer see the tool, and its settings and name stay saved. The switch is unavailable until the Build Agent completes the first setup, and in a versioned application outside a development environment. An agent that is already connected sees a change only after it lists its tools again, which most clients do only when they reconnect.
Endpoint	The address that agents connect to, on the environment's Express host. The page displays it while the switch is ON.

First Setup With the Build Agent

Until a module has a tool, the page displays Set up this tool with the Build Agent instead of the tool's settings. The Build Agent writes the tool's description and each input and output in wording an agent can read. A Creator then reviews and approves the result.

Setting	Description
Set up with Build Agent	A button that opens the Build Agent with a ready-made request and closes the Settings dialog. The button is unavailable while the dialog has unsaved changes. The page then displays Save or discard your other changes first.
Or paste this into the Build Agent	The same request as copyable text, for a Creator who prefers to paste it. The request carries the module's ID.

After the first setup, the page displays the tool's settings for the Creator to review and edit. See Publish API Modules as MCP Tools for the full steps.

Who Can Call It

The Who can call it group lists the Express roles that can call the tool, displayed as badges. A role is listed when it has Write on the module's User Permissions page and its own Default Permission is No Access. The list updates as the User Permissions page changes in the same dialog.

The group displays one of the following alerts when no agent can call the tool or a role is rejected:

Message	Description
Customize RBAC is off, so no agent can call this tool	Customize RBAC for This Module is off. Turn it on in User Permissions, then set the agent's role to Write.
No role can call this tool yet	No role has Write on the module. Set the agent's role to Write in User Permissions. If the environment has no suitable role, create one with Default Permission No Access.
This role can't call this tool or These roles can't call this tool	The alert names each role that MCP rejects and the setting to change. A role is rejected when it has Read-Only on the module, when Designer Access or Full Submission Access is on for the role, or when its Default Permission is not No Access.

Tool Details

The Tool group holds the tool's contract. Agents read the name and description to choose a tool, so both must be clear.

Setting	Description
Tool name	The name agents see. Required. A name uses 1 to 128 characters: letters, digits, underscores, dots, and hyphens (for example, `get_order_status`). The first module to hold a name keeps it. A second tool with the same name is not served until it is renamed, and its page explains why.
Description	The text an agent reads to decide when to use the tool. Required. State what the tool does and when to use it.

Inputs and Outputs

The page reads inputs and outputs from the module's configuration. The request and response panels are where the tool begins. Creators add or remove any field, wherever it sits in the module.

Setting	Description
Inputs	The fields an agent can send. Any component in the module can be an input. The request panel's fields begin in the tool. An agent cannot send a field that the tool does not declare.
Outputs	The fields an agent receives. Any field the module saves to the database can be an output. The response panel's fields begin in the tool.

Each field in the tool displays as a row with the following parts:

Part	Description
Field and description	The field's property ID, a note about what the page read from the module's configuration, and a text box for the field's description. The note names the component, whether the field is in the request panel, the response panel, or elsewhere in the module, and, for an input, the component's own validation rules. Agents read the description, so state what the field means and what to send or expect.
Type	The field's data type. Select Choose a type to open a drop-down. Options include: string, number, integer, boolean, array, object, or ObjectId. Required.
Agent must send	A Required checkbox on input rows only. A field that the module requires stays required while it is in the tool, so its checkbox is locked.
Leave `{name}` out of the tool	The X button at the end of the row. It removes the field from the tool and adds it to the left-out fields below the list.

Fields that the tool leaves out display as buttons below the list. Select a button to add the field to the tool. The buttons are in the following groups:

In the request panel or response panel, left out of the tool: Fields in the panel that the tool does not use.
Elsewhere in the module: Fields outside the panels. For outputs, only fields that the module saves to the database are listed.
Logic and display components: Components that rarely belong in a tool, which a Creator can still add.

Leaving an input out of the tool only restricts MCP. A direct API call to the module can still set the field, so enforce a value that must not change in the module's logic.

Every MCP call runs the module's validation, and validation cannot be switched off. The engine does not validate a field that uses the Hidden component or is hidden on the form. For those fields, only the tool's own checks hold: the field's type and whether the tool marks it required. See MCP Server for API Modules for how failed calls are reported.

Call Limits

Setting	Description
Calls per minute, per agent	The most calls to this tool that one credential can make in a minute. Leave the field blank to use the environment limit, which the page labels Environment default. Accepts whole values of 1 or more.
Calls at once, per agent	The most calls to this tool that one credential can run at once. Leave the field blank to use the environment limit. Accepts whole values of 1 or more.

A tool limit can only be lower than the environment limit. A higher value displays a message such as Calls per minute can be at most 120 here. The page refuses the save until the value is within the environment limit.

Page Messages

The page displays the following messages in addition to the settings above.

Message	Description
Agents can't call this tool right now	The module changed after the tool was published, for example because Customize RBAC for This Module was turned off or an exposed field was removed. The alert also displays when another tool already uses the tool's name, and asks the Creator to rename it. The alert lists the reasons. Fix them, then save.
This application is versioned: the tool goes live when this draft is released.	The application uses versioning, and the tool is saved in the draft. Agents see the tool after the draft's release.
This application is versioned: publish its tools in a development environment, then promote.	The application uses versioning, and the environment is not a development environment. The switch is unavailable.
Complete the MCP Settings page to save.	Displays beside Save while the tool's name, description, or a field's type is missing or invalid.
Mark input “`{name}`” required: the module requires it.	The module requires the field, so the tool must mark it Required.
The module requires “`{name}`”, which the tool leaves out	The tool omits a field that the module requires. Unless the module's logic sets the field, calls run without it or fail.
Use 1–128 letters, digits, underscores, dots or hyphens for the tool name.	The tool name is empty or uses other characters.
Describe what the tool does — the agent chooses tools by this.	The description is empty.
Choose a type for input “`{name}`”. or Choose a type for output “`{name}`”.	A field in the tool has no type. Select one.
Turn on Customize RBAC for This Module (User Permissions) before making it a tool.	Customize RBAC for This Module is off.
Calls per minute can be at most `{number}` here. or Calls at once can be at most `{number}` here.	The limit is higher than the environment limit.
Couldn't load this module's MCP Settings	The page could not load the tool. If the module is a tool, agents can still call it, and saving leaves the tool unchanged. Close Settings and reopen it to try again.

The page also displays cautions as a list below the tool. A caution does not block the save.

Caution	Description
Not marked remote-execute-only, so browsers can load it too.	The module is not set to Server-side Execution Only, so browsers can also load it.
Executes another module, so a call reaches further than this one.	The module runs another module, so a call costs more than the module alone. Consider a lower limit for the tool.

Changelog

Date	Change
2026-11-12	Added a See Also link to MCP in Unqork.
2026-11-05	Initial draft for 9.2.