why is checkout p95 up since the deploy?
It reads the metrics and the traces itself.
Latency from the metrics, the cause from the slowest spans.
Your coding agent, reading the same telemetry you do.
Connect Claude Code, Codex or Cursor and the agent queries metrics, logs and traces the way you do. There is no API key involved, it never exceeds the role of the person who approved it, nothing it can do deletes anything, and every change lands in the audit trail.
Included on the free tier: 20,000 active series, 50 GB of logs and 50 GB of traces a month. No card.
Claude Code · investigating
why did checkout p95 jump at 14:20?
INFOreading checkout p95 over the last 6 hoursstep 1 of 4
INFOp95 2.14 s, up from 640 ms at 14:19step 2 of 4
WARNchat gpt-4o span is 78% of the requeststep 3 of 4
INFOfinish_reason=length on 142 calls this hourstep 4 of 4
What you get
Included on every plan, including the free tier.
Queries the agent ran · this session
QUERYhistogram_quantile(0.95, … checkout …[5m])PromQL
QUERY{ resource.service.name = "checkout" && duration > 2s }TraceQL
QUERY{service_name="checkout"} |= "length"LogQL
INFO20 traces, 142 log lines readdone
Same APIs as your dashboards
The agent reads through the query APIs you use, not a copy of the data.
How it gets there
Three steps, then you can just ask.
No token in a config file. The connection signs in through the browser, as a person.
Add the connection from the portal and sign in through the browser.
Tick the capability groups the agent may use.
Ask your coding agent about your telemetry in its own chat.
# 01 · add the connection, then sign in through the browser
claude mcp add xscaler <your-xscaler-connection-url>opening browser… signed in as [email protected]
# 02 · capability groups ticked in the portal
read metrics · read logs · read traces
you "why is checkout p95 up since the deploy?"
agent read metrics, traces
answer "chat gpt-4o is 78% of the request"
What you can ask
What you would actually type.
Plain questions in your coding agent's chat. It picks the queries and runs them itself.
find the error in payments-api this hour
Logs, filtered to the window that matters.
A LogQL query scoped to the service and the hour.
which label is growing our series count?
Cardinality answered from your own tenant.
Series ranked by label, from the data you are billed on.
Questions
What security reviewers ask about this.
Something not here? Talk to an engineer who works on the platform.
Book thirty minutesA connection acts as the person who approved it and never exceeds that person's role. They tick which capabilities it may use, nothing it can do deletes anything, and every change is attributed in the audit trail.
Start on the free tier.
20,000 active series, 50 GB of logs and 50 GB of traces a month, 90-day retention, and the whole of Insights.