Skip to main content
Custom MCP Tools give you precise control over agent behavior by codifying specific actions into reusable, deterministic steps that execute exactly as defined—every time. You create custom tools within a toolkit and then provide the endpoint to the toolkit to your AI agent. For example, you can create an AI agent using Jira custom tools: the agent finds all unassigned high-priority Jira tickets and assigns them to the appropriate team members. You define the Jira custom tools in a toolkit. You then provide the AI agent with the toolkit endpoint containing the custom tools. Access the Toolkits page by selecting AI > Toolkits in the Connect AI navigation menu.
List of toolkits
The Toolkits home page contains the following:
  • The name of the available toolkits.
  • The number of custom tools each toolkit contains.
  • An icon of the data sources contained within each toolkit.
  • A Copy URL button for copying the URL to the AI agent.
  • A menu icon enabling you to edit, copy, or delete the toolkit.

​
Prerequisites

  • MCP toolkits are available to Growth+ and Connect AI Embed customers only.
  • Only Admin users can create and configure custom tools.
  • To add connection tools, you need to set up one or more connections. See Sources for more information.
  • To add workspace tools, you need to set up one or more workspaces. See Workspaces for more information.

​
Custom Tool Types

There are two main tool types you can add to your toolkit: connection tools and workspace tools. In the edit screen of the toolkit, connection tools are represented by an icon of the tool and workspace tools are represented by a folder.
Edit toolkit page
  • Connection Tools–these tools apply to one or more connections. Within the connection tool, you can activate universal tools and source tools. You can also create custom tools for repeatable SQL queries and operations. Note that once connection(s) are assigned to a connection tool, they cannot be assigned to another connection tool.
  • Workspace tools–these are tools that perform an action specific to the assets in your workspace. Within the workspace tool, you can activate universal tools. You can also create custom SQL tools.
Within connection and workspace tools, select the tab to activate the specific tools available for the tool type.
  • Universal Tools–these are data exploration and manipulation operations built into Connect AI. They expose operations such as getSchemas, getTables, and executeProcedure. They are available for both connection and workspace tools. If you activate the tool, the tool includes every asset in the connection(s) or every asset in the workspace. Click an individual tool to view the description, annotations, and parameters of the tool. Click the information button for a complete explanation of the annotations. Note that universal tools can have dependencies: enabling one tool may require another tool to be enabled.
  • Source Tools (connection tools only)–you can configure actions such as get, search, and create for each asset in a connection in order to optimize your query. For example, if you have a Jira connection, you can enable tools to search for, create, update, and delete issues. You can edit each tool to add AI instructions. You can also view required and optional parameters.
  • Custom SQL Tools–you can create custom SQL tools for repeatable queries and operations. You can add optional or required parameters that are input to the SQL query.

​
Add a New Toolkit

1
In the main Toolkits page, click Add. An Add Toolkit dialog appears.
2
Enter the name of the toolkit, and click Confirm. The new toolkit appears on the main Toolkits page.
3
Select the toolkit to configure it.
4
Add the Toolkit Name. The MCP Remote Server URL is autogenerated. This is the URL that you copy and paste into your MCP client application.
5
Add Server Instructions to the MCP server, setting general toolkit boundaries such as “Use customer data only. Do not use external data.”
Not all MCP clients support server instructions.
6
Click Save Changes. You can now add custom tools to the toolkit.

​
Add a Custom Tool to a Toolkit

1
Open the toolkit you want to add a tool to.
2
Select whether the tool is a connection or workspace tool.
3
Follow the steps in the wizard to add connection(s) or the workspace where the tool should reside. Click Confirm. The tool appears in your toolkit editor.
4
Select the tool to edit it. The tool editor contains the following tabs: Universal Tools, Source Tools (connection tools only), and Custom Tools.
5
In the Universal Tools tab, enable the universal tools for the connection or workspace. Select a tool to view a description of the tool, the tool’s annotations, and the tool’s parameters. Some tools have dependencies with other tools and must be enabled together.
6
(Connection tools only) In the Source Tools tab, enable the tools to be used for the source(s). Select a tool to view a description of the tool and the tool’s optional and required parameters. If desired, add AI Instructions (MCP Context). This is where you can set boundaries and define the intent of the tool. Here are some guidelines for writing effective AI instructions for a custom tool:
  • Be clear and specific about the tool’s purpose. Define what the tool does and what inputs it accepts.
  • Set clear boundaries. Explicitly state what the tool should and should not do.
  • Test your instructions against a wide variety of inputs.
7
The Custom Tools tab is for advanced users. See Add a Custom SQL Tool for details on how to create custom SQL tools for your toolkit.
8
When you are satisfied with the configuration of the custom tools, click Save Changes.

​
Add a Custom SQL Tool

Custom tools can help you improve response accuracy in your AI agent. To add a custom SQL tool:
1
Within the toolkit, select a connection or workspace tool in which to create the custom tool.
2
In the tool editor, click the Custom Tools tab.
Add a custom tool
3
Click Add. The custom tool editor appears.
4
In the custom SQL editor, you can edit the tool name and add AI instructions. You can set boundaries for the type of data that can be returned.
5
Create a SQL statement. The documentation contains a complete SQL Reference for guidance.
6
Optionally, add parameters to the SQL statement. If you add parameters to the SQL statement, you must add the parameter name in the Parameters section of the tool’s definition.
Custom SQL tool
If you use standard SQL for the SQL statement, any parameters you add must be required parameters or the query will fail. See Use SQL Custom Templates for instructions on creating queries with optional parameters.
7
Click Validate SQL. You can also Open in Data Explorer to see the results of the SQL statement.
8
Click Save Changes to save the custom SQL tool.
9
The tool appears in your toolkit. When you are satisfied with the tool’s configuration, enable the tool.

​
Use SQL Custom Templates

Instead of standard SQL, you can use custom SQL templates to create a SQL statement in the custom SQL tool. Custom templates allow you to have optional parameters, or a combination of optional and required parameters. The following two examples follow the SQL template for Bullhorn CRM for a SELECT statement. Refer to the data model for your data source to obtain the accepted parameters for the data source. The following example shows a custom SQL template with both required and optional parameters: 
SELECT * FROM {{catalog_name}}.BullhornCRM.Candidate
WHERE FirstName = @first_name
AND LastName = @last_name
{% if address %} AND AddressAddress = @address {% endif %}
{% if city %} AND AddressCity = @city {% endif %}
{% if state %} AND AddressState = @state {% endif %}
{% if zip %} AND AddressZip = @zip {% endif %}
{% if country_name %} AND AddresscountryName = @country_name {% endif %}
{% if email %} AND Email1 = @email {% endif %}
{% if mobile_phone %} AND MobilePhone = @mobile_phone {% endif %}
@first_name and @last_name are required. The rest of the parameters are optional. In the query, you can include any of the optional parameters, or none. In this SQL template example, all parameters are optional individually, but the template requires that at least one be provided. 
{% if first_name == blank and last_name == blank and email == blank and mobile_phone == blank and city == blank and state == blank %}
SELECT 'Error: at least one search parameter is required' AS ErrorMessage
{% else %}
SELECT * FROM {{catalog_name}}.BullhornCRM.Candidate
WHERE 1=1
{% if first_name %} AND FirstName = @first_name {% endif %}
{% if last_name %} AND LastName = @last_name {% endif %}
{% if email %} AND Email1 = @email {% endif %}
{% if mobile_phone %} AND MobilePhone = @mobile_phone {% endif %}
{% if city %} AND AddressCity = @city {% endif %}
{% if state %} AND AddressState = @state {% endif %}
{% endif %}
This query succeeds as long as one parameter is included. WHERE 1=1 lets you safely append AND clauses without worrying about whether any conditions come before them.

​
Add the Endpoint to Your Agent

Connect AI provides an endpoint to each toolkit. In the main Toolkits page, click Copy URL for the desired toolkit. If you want to view the actual URL, you can open the toolkit editor page. Copy the MCP Remote Server URL and paste it into your AI agent as a custom connector.