Test an MCP tool to make sure it behaves as expected. When you're satisfied with the results, activate it and attach it to the AI experiences that should use it.
Before you begin
- Step 1: Create a connection for the MCP Server connector
- Step 2: Discover and configure the MCP tool
Test the tool
- From the tool's Test step, enter the required sample inputs.
- If the tool uses Context, make sure the appropriate user or case context is available.
- Select Test.
- Review the result and Run trace.
Review the run trace
Use the run trace to understand how the tool ran and identify any problems that occurred.
Depending on the tool, you can review:
- The request sent to the external MCP server.
- The values provided to the tool.
- Any nested tool calls.
- The response returned by the external server.
- Errors encountered during the run.
If the result isn't what you expect, use the trace to identify whether the issue is with the connection, input values, tool configuration, or response.
Activate the tool
When testing is complete, activate the tool so it can be selected by the AI experiences where it is attached.
- Open the tool.
- Set its status to Active.
- Save the tool.
Attach the tool
Attach the active tool to the experiences that need its capability.
You can use MCP tools with supported:
- AI agents
- Case experiences
- Agentic workflows
Choose where to attach a tool based on its purpose, risk, and the permissions required to use it.
For example:
- A read-only ticket-status tool might be appropriate for an employee-facing AI agent.
- A tool that changes a ticket might require additional controls or be limited to a manager, case experience, or workflow.
Test the tool in the AI experience
After attaching the tool, test it in the experience where employees or other users will use it.
- Open the AI experience where the tool is attached.
- Make a request that should require the tool.
- Confirm that the AI selects the tool when appropriate.
- Check the activity message and final response.
- If the tool changes data, verify the result in the external system using a safe test record.
Troubleshoot common issues
| Problem | What to check |
| Connection or authentication error | Connection URL, credentials, and external server availability |
| Permission error | Permissions associated with the MCP connection |
| Unexpected result | Input mappings and values |
| Poor AI response | Tool description and response configuration |
| AI doesn't select the tool | Tool status, attachment, and description |
| Nested tool call fails | Run trace and authentication for the relevant connection |
| Create or update produces the wrong result | Input mappings, conditions, confirmation, and run trace |