Conversational IaC with AWS MCP Servers

What if you could vibe code & engineer your serverless solutions on AWS using dedicated MCP Servers? Just give AI agents access to your AWS accounts. What could possibly go wrong?

Discover how to combine Kiro CLI custom agents, AWS CDK, and AWS MCP Servers to rapidly develop serverless architectures through natural language prompts, and how to do it in a controlled, secure manner.

Transcript

1. Conversational IaC with AWS MCP Servers Jan 28, 2026
2. Playlist 1. From template to dialogue 2. Model Context Protocol 3. AWS MCP Servers 4. Functionless Address Book 5. Demo time! 6. Important security aspects 7. Key takeaways 8. Useful resources 2
3. Infrastructure used to be declared, 
 then coded, then generated.
 Now it can be simply discussed. 3
4. From template to dialogue 4
5. Evolution of Infrastructure as Code 5 Declarative
 IaC Programmatic
 IaC Generative
 IaC Conversational
 IaC Static templates define resources Code generates static templates From prompt to templates or code From dialogue
 to deployment CloudFormation 
 TF / OpenTofu AWS CDK 
 Pulumi / SST Q Developer 
 GitHub Copilot Claude Code CLI 
 Kiro CLI
6. Model Context Protocol 6
7. Model Context Protocol Introduced by Anthropic in Nov 2024, donated to Linux Foundation in Dec 2025 7 Open standard for providing external
 context to LLMs Standardized way how
 AI models connect to 
 data sources & tools Think of USB-C
 for AI-powered applications MCP Servers expose
 data & tools through standard protocol MCP Clients are
 AI agents that connect
 to these servers Universal Protocol replacing fragmented integrations
8. AWS MCP Servers 8
9. AWS MCP Servers AWS IAM AWS IaC AWS Network AWS CloudFormation AWS CDK AWS Terraform AWS Diagram AWS Lambda AWS DynamoDB AWS Core
 dynamic proxy AWS MCP Server
 managed remote AWS Cost Explorer
10. 10 Global agents configuration: ~/.kiro/agents/{agent}.json ~/.kiro/prompts/{prompt}.md ~/.kiro/settings/mcp.json •General-purpose agents across projects •Personal productivity agents •Agents without project-specific context •Commonly used development workflows • Project-specific configurations • Agents with access to selected tools • Environments with unique requirements • Sharing agent configurations via GIT Kiro CLI Agents Local agents configuration: .kiro/agents/{agent}.json .kiro/prompts/{prompt}.md .kiro/settings/mcp.json
11. 11 $ kiro-cli --agent cdk-buddy > /agent [list|swap|generate] > /model > /mcp [list|add] > /tools > /context [show|add] > /prompts [list|get|create] > /chat [save|load|resume] > /checkpoint [list|init|restore] $ kiro-cli agent list $ kiro-cli agent generate $ kiro-cli --agent {name} 
 $ kiro-cli chat --list-sessions $ kiro-cli chat --save $ kiro-cli chat --resume Kiro CLI Agents
12. Functionless Address Book 12
13. Functionless Address Book 13 AWS CDK
 TypeScript Amazon
 DynamoDB AWS Step Functions Region Update Address Book Address Book
 Table Contact Item Event Rule New Contact Event Account Amazon
 EventBridge
14. Custom agent resources 14 PROJECT.md Context about the project: AWS CDK App with TypeScript, AWS CLI profile & region for deployments TASKS.md Detailed instructions how to add AWS resources to the CDK stacks, defined as a list of tasks in sequential order AGENTS.md README for coding agents: provides precise, agent-focused guidance, instructions, extra context, etc
15. Demo time! 15
16. Important security aspects 16
17. Security considerations for MCP 17 Trust and Verification Only install servers from trusted sources Treat server installation like sudo access Check often for security updates 
 Access Control Apply principles of least-privilege (IAM) Restrict access to file system & network Use trust permissions for tools (y/t/n), keep human-in-the-loop Credentials Handling Never hardcode API keys, tokens, etc Use env variables for sensitive data Store credentials securely Use short-lived credentials if possible 
 Sandboxing Run local servers in Docker containers Limit access to required AWS services and shell commands
18. Key takeaways 18
19. Key takeaways 19 Model Context Protocol USB-C for AI-powered applications MCP Servers to access data & tools AWS MCP Servers to interact with AWS 
 Conversational IaC Vibe coding your infrastructure Fine for prototyping, proofs of concept Produced IaC can be production ready Prompt/context engineering takes time Future of IaC Conversational development Spec-driven development (SDD) GitHub Spec Kit Claude Code Plan Kiro IDE Specs Infrastructure as Specs? Kiro IDE Specs Infrastructure from Intent? Spacelift Intent
20. Useful resources 20
21. Useful resources 21 • https://modelcontextprotocol.io/ • https://docs.aws.amazon.com/aws-mcp/ • https://awslabs.github.io/mcp • https://kiro.dev/cli
 • https://serverless.ninja/conversational-infra-as-code • https://speakerdeck.com/conversational-infra-as-code • https://github.com/ServerlessNinja/aws-mcp-addressbook
22. Thank you! Let's keep in touch! 22 https://linkedin.com/in/jakubgaj https://builder.aws.com/community/@jakgaj https://github.com/ServerlessNinja https://serverless.ninja