# Monitor Agent Activity

> For the complete documentation index, see [llms.txt](https://docs.redpanda.com/llms.txt). Component-specific: [agentic-data-plane-full.txt](https://docs.redpanda.com/agentic-data-plane-full.txt)

---
title: Monitor Agent Activity
latest-operator-version: v26.2.6
latest-console-tag: v3.14.0
latest-connect-version: 4.117.0
latest-redpanda-tag: v26.2.7
docname: monitor-agents
page-component-name: agentic-data-plane
page-version: master
page-component-version: master
page-component-title: Agentic Data Plane
page-relative-src-path: monitor-agents.adoc
page-edit-url: https://github.com/redpanda-data/adp-docs/edit/main/modules/monitor/pages/monitor-agents.adoc
description: Monitor agent execution, inspect available MCP tools, analyze conversation history, track token usage, and debug issues using Playground and Transcripts.
page-topic-type: how-to
personas: agent_builder, platform_engineer, domain_evaluator
learning-objective-1: Verify agent behavior and review its MCP capabilities using the <strong>Playground</strong> tab
learning-objective-2: Track token usage and performance metrics
learning-objective-3: Debug agent execution using transcripts
page-git-created-date: "2026-05-28"
page-git-modified-date: "2026-11-02"
page-generated-content: none
---

<!-- Source: https://docs.redpanda.com/agentic-data-plane/monitor/monitor-agents.md -->

Use monitoring to track agent performance, analyze conversation patterns, debug execution issues, and optimize token costs.

After reading this page, you will be able to:

-   Verify agent behavior and review its MCP capabilities using the **Playground** tab

-   Track token usage and performance metrics

-   Debug agent execution using transcripts


For conceptual background on [traces](https://docs.redpanda.com/home/glossary/#trace) and [observability](https://docs.redpanda.com/home/glossary/#observability-o11y), see [How Observability Works](https://docs.redpanda.com/agentic-data-plane/monitor/concepts/).

## [](#prerequisites)Prerequisites

You must have a running agent. If you do not have one, see [Agentic Data Plane Quickstart for Agent Builders](https://docs.redpanda.com/agentic-data-plane/get-started/agent-builder-quickstart/).

## [](#debug-agent-execution-with-transcripts)Debug agent execution with transcripts

An agent’s **Transcripts** tab shows each conversation with timing, errors, and token usage. Use it to debug issues, verify agent behavior, and monitor performance.

### [](#navigate-an-agents-transcripts)Navigate an agent’s transcripts

1.  Open **Agents** in the sidebar and select your agent.

2.  Open the **Transcripts** tab.


The tab lists the agent’s recent conversations, one row per conversation:

-   `Conversation`: The conversation ID, with the conversation title when one exists.

-   `User`: The end user the conversation ran for, or `Unattributed` when no turn recorded one.

-   `Started`: When the conversation began.

-   `Duration`: End-to-end wall-clock time.

-   `Turns`: Number of turns in the conversation.

-   `Status`: `Completed`, `Error`, or `Running`.

-   `Tokens`: Total tokens across the conversation.


Use the search box to match a conversation ID or title, and the status dropdown to narrow the list to `Completed`, `Error`, or `Running` conversations. Use the `Tools` filter and its `Has tool errors` option to keep only the conversations that contain a failed tool call, including the ones that finished as `Completed`. Select the `Started`, `Duration`, `Turns`, or `Tokens` column header to order the whole list by that value. The list is paged, with a rows-per-page selector in the footer. Use the page controls to move through older conversations.

#### [](#conversation-detail)Conversation detail

Click a row to open the conversation. The header shows the conversation ID with a status badge, the start time, duration, and turn count, plus a total-tokens chip. Toggle between two views:

-   **Chat**: Shows the user-visible exchange as a conversation.

-   **Detailed**: Adds per-turn metadata: latency, LLM and tool call counts, input and output token totals, the turn’s context size against the model’s context window, and each tool call with its status and duration.


In **Detailed**, click a tool call to open it and read the **Arguments** the agent sent and the **Result** it got back, plus the error when the call failed. The dialog opens for every tool call and says which of those the transcript didn’t record.

### [](#check-agent-health)Check agent health

Use the **Transcripts** tab to verify your agent is healthy. Recent conversations should show `Completed` status, duration within your expected range, and stable token usage without unexpected growth.

Several warning signs indicate problems. Conversations with `Error` status need investigation. When duration increases over time, your [context window](https://docs.redpanda.com/home/glossary/#context-window) may be growing or tool calls could be slowing down. Many LLM calls for simple requests often signal that the agent is stuck in loops or making unnecessary iterations. If you see no new [transcripts](https://docs.redpanda.com/home/glossary/#transcript), the agent may be stopped or encountering deployment issues.

Pay attention to patterns across multiple conversations. When all recent transcripts show errors, start by checking agent status, MCP server connectivity, and [system prompt](https://docs.redpanda.com/home/glossary/#system-prompt) configuration. A list that alternates between success and error typically points to intermittent tool failures or external API issues. If duration increases steadily over a session, your context window is likely filling up. In the Playground, clear the context to start a fresh session. High token usage combined with relatively few LLM calls usually means tool results are large or your system prompts are verbose.

### [](#debug-with-transcripts)Debug with transcripts

Use the **Transcripts** tab to diagnose specific issues:

If the agent is not responding:

1.  Check the list for recent conversations. If none appear, the agent may be stopped.

2.  Verify agent status in the main **Agents** view.

3.  Look for error transcripts with deployment or initialization failures.


If the agent fails during execution:

1.  Set the status dropdown to `Error` and open the failed conversation.

2.  Switch to the **Detailed** view and find the turn carrying the error.

3.  Check the tool call’s arguments and result for error messages.

4.  Cross-reference with MCP server status.


If performance is slow:

1.  Compare the `Duration` column across recent conversations.

2.  Open a slow conversation in the **Detailed** view and scan per-turn latency to find the bottleneck.

3.  Subtract the turn’s tool time from its latency to see how much belongs to the model calls. See [Find the step that owns the time](https://docs.redpanda.com/agentic-data-plane/monitor/transcripts/#find-the-step-that-owns-the-time).

4.  Verify tool execution time on the nested tool calls.


### [](#track-token-usage-and-costs)Track token usage and costs

View token consumption in the conversation detail view. The **Detailed** view breaks each turn into input tokens (everything sent to the LLM including system prompt, conversation history, and tool results) and output tokens (what the LLM generates in agent responses); the header chip shows the conversation total.

Calculate cost per request:

Cost = (input\_tokens x input\_price) + (output\_tokens x output\_price)

Example, with illustrative prices: 4,302 input tokens and 1,340 output tokens at $0.00000175 per input token and $0.000014 per output token cost $0.026 per request. Check your provider’s current prices for real figures.

For cost optimization strategies, see [Cost calculation](https://docs.redpanda.com/agentic-data-plane/connect/concepts/#cost-calculation).

## [](#test-agent-behavior-with-playground)Test agent behavior with Playground

The **Playground** tab provides real-time conversation testing. Use it to test agent responses interactively and verify behavior before deploying changes.

![Playground test loop. A user sends a prompt, the agent streams responses, artifacts, and MCP tool calls, and Playground shows context and token usage. The user can retry or clear context, while completed runs remain available in transcripts and activity views.](https://docs.redpanda.com/agentic-data-plane/monitor/_images/agent-inspector-test-loop.svg)

Figure 1. Playground closes the loop between a test prompt and production evidence

### [](#access-playground)Access Playground

1.  Open **Agents** in the sidebar.

2.  Click your agent name.

3.  Open the **Playground** tab.

4.  Enter test queries and review responses.

5.  Check the conversation panel to see tool calls.

6.  To start over, click the chat settings control in the composer, then click **Clear context**. See [Reopen a past session](#reopen-a-past-session).


### [](#reopen-a-past-session)Reopen a past session

The Playground saves every conversation with a managed agent as a session, identified by its context ID, so you can return to a test run, read what happened, and carry on from where it stopped.

Click **History** in the Playground header. After a conversation has messages, the same control reads _Session_ followed by the session’s ID, with a copy control and a link icon beside it. Session history lists each session by its title, or by its ID when it has none, along with when it was last updated. Newest sessions appear first. Click **Show more** to load older sessions.

To find one session in a long list, filter the history:

1.  Under Session ID (exact), enter the full session ID.

2.  To filter by date instead, set Updated from and Updated through. Type a date as `YYYY-MM-DD` or pick one from the calendar. Dates use your local timezone, and both ends of the range are inclusive.

3.  Click **Apply filters**. The popover reads _Filters active_ while a filter applies.

4.  Click **Clear filters** to return to the full list.


Click a session to reopen it. The Playground replays the stored transcript, including the files and data sent with each message, and any message you send next continues that session. If an attachment is no longer stored, its place in the transcript says that it could not be displayed. When the task record for a session is unavailable, the transcript still loads, without task, tool, or usage detail, and says so. A very long session shows only its most recent tasks.

The Playground also names the open session in the page address, so reloading the page or opening the address again returns to the same conversation. To share the conversation, click the link icon beside the session ID to copy that address. Opening it loads that session’s history in place of a new conversation, and shows a loading indicator until the history arrives. A new session joins the address when you send its first message, and starting another session takes it out again. If the session can’t be loaded, the conversation says that the session load failed and offers to start a new one.

To delete a session, click the delete icon on its row, then click **Delete** to confirm. Deletion permanently removes the session and its transcript. If you delete the session that is open, the Playground starts a new one.

**New session** in Session history and **Clear context** in **Chat settings** do the same thing. They start a fresh session, and the conversation you leave stays in Session history until you delete it.

A self-managed agent has no Playground tab and no session store.

### [](#send-files-and-structured-data)Send files and structured data

Click **Add attachment**, the paperclip control in the composer, to send more than text with your next message. **Attach files** takes one or more UTF-8 text files (TXT, CSV, Markdown, or JSON) of the types the agent accepts. When the agent accepts none of them, file uploads are unavailable. **Add structured data** opens a dialog for a single JSON object, which the agent receives as a data part alongside whatever you typed. You can also drag files straight onto the composer, which reads _Drop files to attach_ while you hold them over it.

Each attachment appears as a chip above the message box, and you can remove any of them before sending. A file still being read shows _Reading…_, and the send control stays disabled until every file finishes. Attachments must total 10 MiB or less per message. Past that, Playground asks you to remove a file and try again.

A test message doesn’t need text. A file or a structured-data object on its own is enough to send a turn, which is useful for checking how an agent handles an attachment it wasn’t prompted about.

To add structured data, enter one JSON object under Structured data JSON, optionally click **Format JSON** to reindent it, then click **Add data**. The dialog reports both invalid JSON and valid JSON that isn’t an object, and keeps your draft on screen so you can correct it.

### [](#review-mcp-tools-and-subagents)Review MCP tools and subagents

For a managed agent, the Playground composer has a chat settings control, labeled with the model the conversation uses. Click it to open **Chat settings**, which holds the model selector, a **Tools** control under an MCP label, and, when the agent delegates to subagents, a Subagents control that shows how many it has, such as **2 subagents**. Use them to review the capabilities available to the agent before you send a test message.

To review the agent’s direct MCP servers and their tools:

![Playground tool discovery. The Tools control displays direct MCP servers, and the Subagents control displays the agent’s subagents. Direct servers expose their tools to the agent. Each subagent exposes only the tools from the MCP servers configured for that subagent. Authentication and availability affect which tools are visible.](https://docs.redpanda.com/agentic-data-plane/monitor/_images/inspector-mcp-tool-discovery.svg)

Figure 2. Playground shows tool access through direct MCP servers and subagents

1.  Open the agent’s **Playground** tab.

2.  Click the chat settings control in the composer, then click **Tools**.

3.  Under Direct MCP servers, expand a server to see its type, description, and exposed tools. Each tool entry shows its name and description when the server provides one.


The list reflects the agent’s current configuration and the tools each direct MCP server exposes. It does not guarantee that a later tool call succeeds. Server availability, access policy, or an external connection can still affect a call at runtime.

Some remote MCP servers require you to connect your own account before their tools are shown. Expand the server and click **Connect**, complete the authorization in the new browser tab, then return to Playground. The list refreshes when you return. If it does not, click **Refresh tools**.

Use the status shown under a server to resolve common issues:

| Status | What to do |
| --- | --- |
| Authentication required | Click Connect and complete the authorization flow. Protected tools remain hidden until the connection grants the required access. |
| Unavailable | Ask an administrator to confirm that the MCP server exists and is enabled for the agent. |
| Couldn’t load tools | Click Try again. If the error continues, check the MCP server’s availability and authentication settings. |
| No tools exposed by this MCP server | Confirm that the server publishes tools and that the agent is connected to the intended server. |

When the agent delegates to subagents, click the subagents control next to **Tools** in **Chat settings** to review the delegation targets this agent can orchestrate. Each entry lists the subagent’s name, its description, and the MCP servers configured on it.

### [](#outputs)Read the agent’s outputs

Whatever the agent supplies in reply to a test message appears in the conversation itself, alongside the response text.

-   An image the agent returns is shown in place, captioned with its filename and media type.

-   Any other file appears as a row carrying its filename and media type. A file the agent hosted opens in a new tab, and a file it returned inline downloads.

-   Structured data appears as a collapsed **Structured data** block. Expand it to read the JSON.


A file the Playground can’t display says so in place of the file, and asks you to have the agent supply it again over HTTPS. In a reopened session, an attachment that is no longer stored says that instead. See [Reopen a past session](#reopen-a-past-session).

An image the agent links to in its reply text is handled differently from a file it returns. The Playground doesn’t fetch it until you say so. In place of the image you get its description, the address it would load from, and a **Load image** control. For a remote address the placeholder also warns you that loading the image sends a request to that address. That’s what the wait is for, because an address the agent chose can see the request. Approving an address loads that image wherever it appears again in the conversation. An embedded image carries no address and nothing is requested, so the placeholder just names it. An image the Playground can’t load at all, because its address isn’t an `http`, `https`, or embedded image address, says its source isn’t supported and offers no control.

### [](#reply-detail)Read the detail behind a reply

Each tool call the agent makes appears in the conversation as one line that reads _Ran_, _Running_, or _Failed to run_, followed by the tool’s name, with how long the call took beside it. Expand a line to read the call’s input and result, the ID of the call, and a control that copies the tool’s details. A call the agent is still running has no duration until it finishes.

When a model returns a summary of its own reasoning, the conversation carries a **Reasoning summary** line. Expand it to read the summary the agent supplied.

Under each agent reply, click the response details control for the counts behind that turn: its input and output tokens, the time it was recorded, and the ID of the message. A value the model provider didn’t return is left out. Use **Copy message** beside it to copy the reply’s text.

### [](#activity-log)Read the activity log

Click **Activity** in the Playground header to read the work behind the conversation: the tools the agent called, the statuses it reported, and any warnings, errors, or connection events. The log records what happened and takes no action on it.

Entries appear in conversation order. Each one carries a title and a status, plus the time to the millisecond when one was recorded. A tool call also carries the ID of the call and how long it took, or _Duration unavailable_ when no timing was recorded, and an entry with further detail shows it underneath.

The log holds what the agent reported and nothing else, so a missing result doesn’t mean the tool is still running. Until the agent reports something, the log reads _No activity recorded_.

The log covers the conversation you have open in the Playground. For the record that outlives a test run, read the agent’s transcripts. See [See What Your Agent Did](https://docs.redpanda.com/agentic-data-plane/monitor/transcripts/).

### [](#context-window-usage)Token usage and context estimate

As you test, the Playground reports how many tokens the conversation has used, and estimates how much of the model’s context window the latest turn occupies. The indicator sits in the conversation header, beside the session control, under a **Context** label. It shows the context estimate as a percentage with a ring when Redpanda Agentic Data Plane knows the model’s limits, and falls back to the session’s input and output token counts when it doesn’t.

Click the indicator to open the breakdown. A Token usage section reports the Latest input token count, the Model input limit, the Model output limit, and the Context estimate as a percentage. A Reported session usage section reports the Session total and splits it into Input, Cached input, Output, and Reasoning output.

The context estimate and the reported counts answer different questions, so read them separately:

-   The reported counts are measurements the model provider returned for the session. A heading of _Reported session usage (partial)_ means the provider returned some counts but not a complete breakdown, and a session with no counts at all reads _Token usage unavailable._

-   The context estimate is a ratio of the latest input against the model’s published input limit, which is why it carries a tilde. It isn’t live telemetry from the model, and it doesn’t account for anything the provider does to the context that it doesn’t report.


The latest input count includes cached tokens, so the estimate reflects the full context occupancy under prompt caching, not fresh input alone. As a conversation grows and the estimate approaches 100%, clear the context to start a fresh session before responses slow down or the agent starts dropping earlier context.

Agentic Data Plane reads the two model limits from its model catalog. For a model that isn’t in the catalog, such as one served by an OpenAI-compatible provider or a hand-entered model ID, the limits and the context estimate read _Unavailable_, and the indicator reports the reported token counts alone.

### [](#stop-a-run)Stop a run

While the agent is working, the send control in the composer becomes **Cancel**. Click it to stop the run. Playground ends the live view and labels the turn _Canceled_ after the agent confirms the cancellation.

### [](#long-running-tasks)Long-running tasks

The Playground live view streams a test run as it happens. If the stream drops, Playground tries to reconnect. When it can’t reconnect within about five minutes, it stops the live view and shows a Live updates stopped notice. The task may still be running, so check the agent’s **Transcripts** tab later for the result.

### [](#testing-best-practices)Testing best practices

Test your agents systematically by exploring edge cases and potential failure scenarios. Begin with boundary testing. Requests at the edge of agent capabilities verify that scope enforcement works correctly. Error handling becomes clear when you request unavailable data and observe whether the agent degrades gracefully. Even with proper system prompt constraints, testing confirms that your agent responds appropriately to edge cases.

Monitor iteration counts during complex requests to ensure they complete within your configured limits. Ambiguous or vague queries reveal whether the agent asks clarifying questions or makes risky assumptions. Throughout testing, track token usage per request to estimate costs and identify which query patterns consume the most resources.

## [](#next-steps)Next steps

-   [How Observability Works](https://docs.redpanda.com/agentic-data-plane/monitor/concepts/)

-   [Visualize Your Agent Network](https://docs.redpanda.com/agentic-data-plane/monitor/agent-network/)

-   [Troubleshoot Agents](https://docs.redpanda.com/agentic-data-plane/monitor/troubleshoot-ai-agents/)