mcp

This repository is a collection of open-source MCP servers built by AWS Labs. MCP stands for Model Context Protocol, an open standard created by Anthropic that lets AI coding assistants and chatbots connect to external tools and data sources in a consistent way. When you configure one of these servers alongside a tool like Claude, Cursor, Windsurf, or VS Code's AI features, the AI can query AWS documentation, deploy infrastructure, run database queries, and more, all from within a chat interface.

The collection covers a wide range of AWS services. There are servers for looking up official AWS documentation, generating and applying infrastructure-as-code with CDK or CloudFormation, working with serverless and containerized workloads, querying SQL and NoSQL databases, running analytics, managing caching layers, and handling healthcare and life-sciences workflows. Each server is an independent Python package you install and configure separately.

Installing a server typically means running pip install and adding a short JSON configuration block to your AI tool's settings file. The README includes ready-to-paste configuration snippets for Claude Code, Cursor, Windsurf, VS Code, Cline, and Kiro. The servers communicate over standard input and output by default, so they do not open any network port unless you configure a remote transport.

AWS Labs notes that this collection is evolving into something called the Agent Toolkit for AWS, which includes finer-grained IAM permission controls, CloudWatch logging, and built-in accuracy evaluations intended for production deployments. The MCP servers in this repository continue to work and accept contributions; over time the most useful ones are expected to move into the Agent Toolkit.

The full README is longer than what was shown.