Cloud Infrastructure

AWS Gives Agents a Read-Only View of Lambda Failures

By Kaleido Field Staff · September 6, 2026

Citation-ready: AWS's new MCP Server serverless capability is read-only, scoped to the caller's account, and designed to diagnose Lambda functions and connected resources.

What happened and why it matters

A structured incident view can reduce the number of separate evidence-gathering calls. Its read-only boundary leaves operators responsible for deciding whether a proposed change is justified and how to verify recovery.

The dated source record

Primary reference: AWS: MCP Server adds a serverless capability for Lambda. Kaleido Field checked the event date and the article's attributed facts against this source.

The evidence bundle has specific contents

The capability guide describes a combined health check, a seven-day baseline, grouped log exceptions, deployed configuration, and a timeline of changes. Trace summaries require X-Ray tracing to be enabled.

An absent trace should be reported as unavailable evidence. It should not become a clean bill of health. Likewise, a change just before an outage is a candidate explanation until the failure mechanism has been checked.

Keep the proposed fix outside the diagnostic receipt

An incident record should preserve the tool inputs, time window, account and region, returned observations, and the proposed explanation. If an operator approves a fix, record that decision and the exact mutation separately.

After the mutation, compare the same health indicators and configuration fields. A successful deployment status only confirms that a rollout completed; it does not establish that the original user-visible failure has stopped.

Availability and cost need precise wording

AWS says the managed server runs in Northern Virginia and Frankfurt while accessing commercial AWS regions, and that this diagnostic capability has no additional charge. That does not imply the underlying workload, telemetry, or model calls are free.

The related HyperPod control-plane story concerns executing infrastructure operations. This release adds a narrower, read-only incident investigation path.

Evidence boundary

Official release and documentation facts. Token savings are AWS positioning without an independent measured comparison here. The tool's suspected cause is a diagnosis, not causal proof, and read-only diagnostics do not constrain unrelated tools in the same agent session.