Skip to navigation
Agent Telemetry

Agent Telemetry

What you can observe once a tool is live, from per-execution logs to assistant-wide analytics.

View as Markdown

Once a plugin, MCP server, or system trigger is live, four surfaces tell you what it did. Logs record each execution step by step and are where to start when a run fails. MCP tool call logs are the subset of those entries that show what the assistant sent to an MCP tool and what came back. AI Assistant Insights aggregates adoption and plugin performance across users. The Data API exports the underlying interaction records so you can build dashboards in your own BI tools.

Logs

Filter plugin executions by time range, Root UUID, requestor, log level, log type, and response code.

MCP Tool Call Logs

Confirm which server and tool ran, inspect arguments and the model-visible response, and diagnose failures or latency.

AI Assistant Insights

Interactions, user adoption, and the success rate of each plugin call, with in-chat feedback from end users.

Data API

A REST API that exports conversations, interactions, invoked plugins, and users for your own analytics.

Logs

Open the Logs app from the App Switcher at my.moveworks.com. The log viewer filters are:

FilterWhat it does
Time rangeBounds the search. Defaults to the last 6 hours and can go back a maximum of 14 days. Timestamps indicate approximately when activity occurred, so use the range to locate logs rather than to profile latency.
Root UUIDThe UUID that ties an entire plugin execution together.
RequestorThe email of the user who triggered the plugin.
Log LevelLOG_LEVEL_DEBUG by default; LOG_LEVEL_ERROR when an error occurred.
Log TypeThe stage of execution, such as compound_action.trigger, compound_action.step.execute, or action.http.trigger.
Response CodeFor action.http logs only: filter by the HTTP response code, such as 200 or 404.

The Log Type Reference lists the fields in each log type, including the webhook listener stages. Log Data Security explains redaction: Standard Log Redaction applies to all logs and cannot be turned off, Strict Log Redaction can be set per plugin or MCP server from its Launch Configuration, and Permissive Log Redaction is available only when your super admin enables it. When you ask ToolSmith to fix a plugin, it reads these execution logs to find what went wrong.

MCP Tool Call Logs

Every MCP tool call creates an entry in the Logs app. Find them by the conversation’s Root UUID, by selecting the MCP server in the Plugin filter, or by selecting mcp.tool.trigger in the Log Type filter. Each entry can include the server, tool name, arguments, the response shown to the model, execution status, latency, and error message; arguments, response, and error are each limited to 32 KB. View MCP Tool Call Logs maps what you see in an entry to what to check next.

Assistant Analytics

AI Assistant Insights covers every type of interaction with the assistant (free-form text, button clicks, link clicks, and UI form submissions), user adoption from active user counts to popular plugins, and performance: the success rate of each plugin call, the causes of unsuccessful interactions, and in-chat feedback from end users. Interactions from users tagged as testers are excluded.

Data API

The Data API is a REST API for ingesting conversation and interaction data into your own systems. It exposes all interactions, the plugins invoked by the LLM, the resources used to generate responses, and the users who have access to the assistant. Use it when AI Assistant Insights does not answer the question and you want to build custom dashboards in a BI tool.