> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.fabricgateway.ai/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.fabricgateway.ai/_mcp/server.

# Monitor gateway traffic with analytics and traces

Fabric Gateway records every request that passes through it, so you can monitor traffic without instrumenting your own services. Use **Analytics** to see aggregated metrics like cost, latency, and error rate across a time window, and use **Traces** to inspect individual requests down to the plugin and upstream level.

Learn how to read the dashboards and how to find and inspect a trace.

## Understanding analytics

**Analytics** enables you to monitor **LLM** and **MCP** traffic within these time windows: one hour, 24 hours, seven days, or 30 days.

### View LLM traffic analytics

To view LLM analytics, do the following:

1. In the sidebar, select **Analytics** under **Observability**.
2. Select the **LLM** tab.
3. Select a time window.

The LLM tab shows the following:

| Section        | What it shows                                                                                    |
| -------------- | ------------------------------------------------------------------------------------------------ |
| Summary cards  | Total requests, total tokens, estimated cost, and error rate for the selected window.            |
| Cost over time | A chart of estimated spend broken out by model.                                                  |
| Top models     | A table of models ranked by estimated spend.                                                     |
| Top consumers  | A table of consumer identities ranked by estimated spend, with requests and tokens per consumer. |

### View MCP traffic analytics

To view MCP analytics, do the following:

1. In the sidebar, select **Analytics** under **Observability**.
2. Select the **MCP** tab.
3. Select a time window.

The MCP tab shows the following:

| Section       | What it shows                                                                                                                                 |
| ------------- | --------------------------------------------------------------------------------------------------------------------------------------------- |
| Summary cards | Total tool calls, error rate, P95 latency, and active consumer count for the selected window.                                                 |
| Activity      | A requests-per-minute chart broken out by tools, resources, and prompts.                                                                      |
| Top tools     | A table of MCP tools (for example, `search`, `integrate`, `execute_nrql_query`) ranked by requests, with error rate and P95 latency per tool. |
| Top consumers | A table of authenticated identities ranked by request volume, with error rate and P95 latency per consumer.                                   |

## Understanding traces

**Traces**, under **Observability** in the sidebar, lists every distributed trace recorded by the gateway. Use Traces when you need to debug a specific request rather than look at aggregate metrics.

### Find a trace

To find a trace, do the following:

1. In the sidebar, select **Traces** under **Observability**.
2. Optionally, select a time range: **24H**, **7D**, **30D**, or **90D**.
3. Optionally, add one or more filters: **Protocol**, **Route**, **Service**, **User**, or **Status**.

The traces list shows summary cards for total traces, P95 latency, and error rate over the selected range, followed by a table with the following columns: **Trace ID**, **Timestamp**, **Protocol** (HTTP, LLM, or MCP), **Operation**, **Route**, **Service**, **User**, and **Status**.

### Inspect a trace

Select a row in the traces table to open its span view.

Each trace is made up of one or more spans — for example, an HTTP request through an imported service has a gateway span and a nested span for the call to the upstream. Selecting a span shows its attributes, including:

* Plugin execution time and invocation count
* Upstream attempt count, body-read count, and duration
* Route match type and route name
* Tenant ID
* HTTP method and status code
* The originating user agent

To learn what the plugin-related attributes in a span correspond to and reduce errors or latency shown in analytics or traces, see [Control and transform gateway traffic with plugins](/fabric/plugins/).