AWS re:Invent 2025 - An unexpected journey building AWS MCP Servers (OPN401)

AWS re:Invent 2025 - An Unexpected Journey Building AWS MCP Servers

Overview and Motivation

• Presenters Paul Vincent and Lethal Sedun are principal architects at AWS
• They developed the AWS MCP (Model Context Protocol) servers to accelerate cloud development and prototyping
• The goal was to provide AWS-specific "kung fu" to AI coding assistants, allowing them to quickly build quality cloud solutions

Community Adoption and Impact

• The MCP servers have seen tremendous community adoption:
  • Over 7,400 GitHub stars
  • Thousands of forks and community contributions
  • 60+ specialized MCP servers across use cases
  • Millions of downloads
• Developers praise the MCP servers for providing the "cloud dev experience" they were looking for:
  • "This isn't full AWS automation. It's how cloud dev should feel."
  • Allows developers to focus on building rather than guessing AWS services, policies, etc.

Origins and Goals

• The presenters' team at AWS does rapid cloud prototyping for customers
• They found developers were spending too much time on AWS knowledge transfer and lookup, rather than actual development
• The MCP servers were created to "jack in" AWS knowledge and capabilities into AI coding assistants
• This allows for much faster prototyping and development cycles, going from 6-8 weeks down to 5 days

MCP Server Capabilities

• The MCP servers are organized into 3 main categories:
  1. Documentation and Knowledge: Provides direct access to AWS documentation for developers
  2. Workload Specialization: Servers for specific AWS services like EKS, Lambda, etc.
  3. Developer Specialization: Servers for infrastructure as code tools like CDK and Terraform
• Other specialized servers handle pricing, diagramming, front-end development, and more

Design Principles and Choices

1. Multi-Server Separation: MCP servers are designed as isolated, composable tools rather than a single monolithic server
   • Allows developers to only load the specific servers/context they need
2. Multi-Server Composability: The "core" MCP server dynamically binds other specialized servers based on user roles/profiles
   • Avoids the need to load all 60+ servers at once
3. No One-to-One API Wrappers: MCP servers provide higher-level, multi-API tools rather than simple 1:1 API wrappers
   • Enables more natural, workflow-oriented interactions for agents

Deep Dive: Core MCP Server

• The core MCP server is responsible for dynamically binding other specialized servers based on user roles/profiles
• It uses environment variables to determine which role-based profile to load, then imports and proxies the corresponding servers
• This allows developers to control the context and capabilities exposed to their AI agents

Deep Dive: Billing and Cost Management (BCM) MCP Server

• Consolidates multiple AWS services related to billing, cost optimization, and cost analysis into a single server
• Uses an in-memory SQL database to store and query cost and usage data, providing a more developer-friendly interface
• Includes specialized tools for working with AWS Cost and Usage Reports, Storage Lens, and other billing-related services

Key Takeaways

• The MCP servers are designed to accelerate cloud development by providing AI coding assistants with deep AWS knowledge and capabilities
• The modular, composable architecture allows developers to control the context exposed to their agents
• The servers go beyond simple API wrappers, providing higher-level tools and workflows for common cloud development tasks
• The open-source nature encourages community contributions and customization for internal use cases
• AWS has seen significant productivity gains from using the MCP servers in their own prototyping and development work

Getting Involved

• The MCP servers are open-source and available on GitHub
• Developers are encouraged to contribute, provide feedback, and adapt the servers for their own internal use cases