Aws Mcp

AWS MCP is an open source Model Context Protocol server that lets an AI assistant such as Claude query and manage an Amazon Web Services account in plain language. Once connected, a user can ask for things like a list of EC2 instances, S3 buckets with their sizes, Lambda functions deployed in a given region, or ECS clusters and their services, and the assistant calls the AWS APIs on their behalf. The author describes it as an alternative to Amazon Q. It is built for developers and cloud engineers who already work with AWS from the command line and want to inspect or change resources during a conversation instead of switching to the console. It supports multiple AWS profiles and SSO authentication, works across regions, and runs locally using the credentials already configured in the user's AWS directory, so credentials are not sent to any external service. Setup requires Node.js and Claude Desktop: the user clones the repository, installs dependencies with npm or pnpm, and adds an entry to the Claude Desktop configuration file. Instructions are also provided for machines that manage Node.js versions with nvm, along with log locations for troubleshooting. The project is free to use. It is published as source code on GitHub and there is no paid tier or hosted service. The listed website is the personal site of the author, Rafal Wilinski, an independent AI consultant who also built the DynamoDB client Dynobase and the ChatWithCloud command line tool; the server itself lives in his GitHub repository. Because it can modify resources as well as read them, it should be pointed at an AWS profile with appropriately limited permissions.