AWS MCP Server Adds Lambda Troubleshooting for Coding Agents

What AWS shipped

AWS has added a serverless capability to the AWS Model Context Protocol Server, its managed service that gives AI coding agents secure access to AWS resources. The update lets agents such as Claude Code and Kiro diagnose problems in running Lambda functions and the resources connected to them. 1

The scope goes well beyond the function itself. An agent using the new capability can inspect a Lambda function along with its links to API Gateway, EventBridge, S3, DynamoDB, SNS, SQS and Step Functions. 1 From there it can:

• compare error signals against a seven-day baseline to isolate what changed
• flag recurring errors to show trends
• pull the deployed configuration of the function and its dependencies
• build a timeline of recent changes
• analyze latency across connected services 1

AWS says the agent receives this data in a single call. 1 That detail matters more than it might seem.

The MCP Server can be installed on its own or through the Agent Toolkit for AWS. 1

The packaging: plugins and a toolkit

The troubleshooting capability is part of a broader effort to make AWS a first-class target for AI-assisted development.

AWS also announced the Agent Plugin for AWS Serverless, which works with Kiro, Claude Code and Cursor. 2 Agent plugins bundle several things into one modular unit [2]:

• skills
• sub-agents
• hooks
• MCP servers

The serverless plugin loads relevant guidance as needed across the development lifecycle. That covers Lambda integrations with EventBridge, Kinesis and Step Functions, plus infrastructure-as-code setup through SAM and CDK, CI/CD pipelines and local testing. 2

The AWS Labs GitHub repository shows where this is heading. It describes the Agent Toolkit for AWS as the successor to the MCP servers, plugins and skills previously published under AWS Labs. 3 AWS recommends the toolkit for teams shipping production software with coding agents or building agents for their own customers. 3 The toolkit adds several governance features [3]:

• IAM condition keys that distinguish agent actions from human ones
• visibility in CloudWatch and CloudTrail
• skills that have been evaluated for accuracy and effectiveness

The repository currently lists plugins for serverless development and for SageMaker AI as available. 3

Overlap with Serverless Framework

AWS is not alone here. Serverless Framework v4.43.0 introduced commands that help AI coding agents set up, build, deploy and debug AWS services. Its pitch is explicitly about avoiding the token cost of querying AWS one call at a time. 4 Earlier releases let developers deploy MCP servers to Lambda with a few lines of YAML. 4 The framework has also added support for Lambda MicroVM sandboxes and Bedrock AgentCore. 4

The two efforts address the same bottleneck. A coding agent that has to stitch together logs, configurations and metrics through dozens of separate API calls burns context and money. It is also more likely to lose the thread. Both vendors are trying to pre-aggregate the diagnostic picture so the agent can reason over it in one pass.

The announcements do not frame this as a direct response to Serverless Framework. Still, the overlap in goals and timing is hard to miss, and the toolkit's governance features are an area where a platform owner has a natural edge.

Why it matters

The more significant shift is not the troubleshooting feature itself. It is AWS consolidating a scattered set of experimental MCP servers into a supported, auditable product. 3

IAM condition keys that separate agent actions from human ones address a real concern for security teams. Many organizations have hesitated to give autonomous agents credentials because their actions blur into a developer's audit trail. Native CloudTrail and CloudWatch visibility gives those teams something concrete to evaluate. 3

The seven-day baseline comparison also reflects how operations work in practice. Most production incidents trace back to a recent change, whether a deployment, a configuration edit or a shift in an upstream service. An agent that can line up a change timeline against an error spike is doing what an on-call engineer would do first. 1

The takeaway

AWS is positioning itself as the default substrate for agent-driven cloud work. It offers a managed MCP endpoint, curated plugins and IAM-level controls, all native to the platform. Third-party tools like Serverless Framework still compete on developer ergonomics and multi-feature convenience. 4 But on governance and deep service integration, AWS holds structural advantages that are difficult to replicate from outside.

The open question is quality. The value of single-call diagnostics depends on whether agents draw correct conclusions from the data they receive. AWS's note that toolkit skills are evaluated for accuracy suggests the company knows that is where trust will be won or lost. 3