Query telemetry with AI Chat

Overview

Telemetry Query Model Context Protocol (MCP) Tools let you ask questions in AI Chat about telemetry that may not be covered by a dedicated tool or dashboard. The assistant checks which telemetry is available, builds a query for the relevant data source, and uses the results to answer your question.

These tools extend the existing MCP tools in AI Chat. The assistant uses a dedicated tool when it can answer your question directly.

Note

  • Telemetry Query MCP Tools are an early access feature.
  • To enable or disable them, navigate to Settings > About > Manage features and toggle AI Chat Telemetry Query Tools.
  • The tools are enabled by default in new installations, but an administrator can change the setting.

Before you begin

Make sure that:

  • AI Chat is available and configured for your F5 Insight account.
  • The early access toggle for AI Chat Telemetry Query Tools is enabled.
  • F5 Insight has collected the telemetry needed to answer your question.

The MCP tools can query the following data sources for BIG-IP telemetry:

Data source What it contains Accessible tables
VictoriaMetrics Time-series metrics (counts, rates, trends) All collected metric modules
ClickHouse Structured log and event records irule_logs, sslo_logs

The tools can’t access F5 Insight administrative settings, internal databases, or sensitive data such as passwords and licenses.

Ask a telemetry question

Ask your question in AI Chat. Include details that help identify the data you want, such as:

  • The device, devices, or data center
  • The time period
  • The data you’re looking for (such as CPU usage, logs, or certificate details)
  • Any grouping or comparison you want (such as by virtual server or device)

If it’s unclear whether your question needs metrics or log data, the assistant may ask you to clarify. If F5 Insight doesn’t collect the requested data, the assistant indicates that the data is unavailable rather than generating an answer.

How the assistant queries telemetry

The assistant uses a discovery step before querying each data source. It only discovers the source relevant to your question.

Query metrics in VictoriaMetrics

For metrics, the assistant:

  1. Checks the available metric modules and groups.
  2. Looks up the metrics and labels available for the relevant modules.
  3. Builds and runs a PromQL query using that metadata.

Use this path for time-series data, such as counts, rates, or trends over time.

Query data from ClickHouse

For logs, the assistant:

  1. Checks which ClickHouse tables are available and reads their column information.
  2. Builds a SQL query using the discovered table and column names.
  3. Runs the query and uses the returned rows to answer.

Use this path for structured log or event records. The tables available depend on the F5 Insight deployment.

The deployment can restrict the available tables. The tool can’t add ClickHouse tables beyond its supported set.

Query limits

Limits help keep queries and results manageable. The defaults can be changed by the F5 Insight deployment.

Data source Default limits
VictoriaMetrics Time-range queries can cover up to 30 days. Results are limited to 100 series and 100 samples per series.
ClickHouse Queries must use a read-only SELECT or WITH statement, include a LIMIT of no more than 100, and return no more than 100 rows.

If a result reaches a limit, it may be partial. The assistant explains this and asks you to narrow the time period or scope. Verify whether a capped result represents the complete data set before relying on it.

Data handling and privacy

Query restrictions

  • The ClickHouse query tool accepts one read-only query at a time.
  • It rejects write or schema-change statements, multiple statements, and references to tables outside the allowed set.

Privacy considerations

  • The available ClickHouse columns may include source or client IP addresses. The query tool doesn’t automatically remove these columns from results.

Audit logging

  • F5 Insight records AI tool activity through its audit system when audit context is available.
  • If the audit system can’t record the tool call, that tool execution is blocked.

Understand the answer

The assistant bases its answer on the data returned by the query. Results can be empty because:

  • No matching telemetry was found.
  • The requested information isn’t collected.
  • The query scope or time period didn’t include matching data.

When a query fails or a data source is unavailable, the assistant reports that instead of presenting generated results. If the result is partial or capped, review the stated limit and narrow the request before relying on the answer.

You can ask for a chart when that would help. AI Chat can use its existing visualization tools to present supported results.

Example questions

Try these questions in AI Chat to explore your telemetry:

  • “Which BIG-IP devices have sustained high CPU or memory usage over the last 30 minutes across all devices?”
  • “Show SSL certificate expiry details for device <DEVICE_NAME>.”
  • “Get the top 10 processes consuming the most CPU in the last hour for device <DEVICE_NAME>.”
  • “Show the most accessed URIs with 400 and 500 error codes across all devices in the last 24 hours.”
  • “Show the most accessed URIs with 200 status codes across all devices in the last 24 hours.”

Replace <DEVICE_NAME> with the name of a BIG-IP device in your fleet.

Known limitations

Telemetry Query MCP Tools:

  • Can only answer questions supported by collected telemetry and the enabled data sources.
  • Don’t provide access to administrative settings or data from PostgreSQL or Vault.
  • Don’t guarantee complete results when a query reaches a configured limit.
  • Don’t replace curated dashboards, dedicated AI tools, or deeper analysis workflows.

Check the data source, time period, device scope, and any stated limits before relying on an answer.