Resources for SVS324: Building agentic apps with serverless: Lambda as your agent's toolbox
Agents need tools to retrieve data, take actions and reach external systems, and designing AWS Lambda functions as agent tools takes different thinking than building API backends. This chalk talk worked through a multi-tenant commerce platform with two patterns: exposing existing Lambda functions as MCP tools through Amazon Bedrock AgentCore Gateway, and running an MCP server on AWS Lambda Web Functions. It covered tool design and identity propagation from the caller's token to the tool.
This page has the slides and the links for each part of the talk.
This page has the slides and the links for each part of the talk.
AgentCore Gateway: Lambda functions and MCP servers as targets
Expose existing functions as tools
Put Amazon Bedrock AgentCore Gateway in front of the Lambda functions that already back your APIs, or in front of your own MCP server, and the Gateway handles the MCP protocol.
MCP servers on AWS Lambda Web Functions
Docs
Run an MCP SDK server on Lambda Web Functions when you offer tools, resources and prompts to other people's agents, with AgentCore Gateway in front.
Well-Architected Agentic AI Lens
Designing the tools
Best practices for tool design: strict schemas, validated input and output, and meta-tools that cut reasoning loops.
Identity from token to tool
Multi-tenant isolation
The tenant comes from a validated token, never from a model argument. AgentCore Gateway validates inbound identity, applies policy and interceptors, and handles outbound credentials.
Presented by:
MCP for agent tools
The protocol
MCP is how remote tools reach agents. A tool holds the business logic, and one MCP server can host many tools.