AWS MCP Servers
About This Server
AWS's official open-source suite of MCP servers lets AI agents work with the AWS cloud in natural language — read docs, manage infrastructure, drive CDK/Terraform, access S3/Lambda/EKS and dozens of purpose-built servers, with a core server that dynamically proxies other AWS MCPs on demand.
Installation
uvx awslabs.core-mcp-server@latestFrequently Asked Questions
Is AWS MCP one server or a suite?
A whole suite. The awslabs/mcp repo contains dozens of purpose-built MCP servers split by service (docs, CDK, Terraform, S3, Lambda, EKS, cost analysis, etc.); you install only the ones you need, and the core-mcp-server can act as a dynamic proxy that loads other AWS MCPs on demand via environment variables.
What prerequisites does AWS MCP need?
The servers are written in Python and run via uv/uvx (e.g. uvx awslabs.core-mcp-server@latest); you need uv plus Python 3.10+ installed, and AWS credentials with appropriate permissions configured (e.g. ~/.aws or environment variables). It operates on resources in your own AWS account.
Is AWS MCP free? Will it incur cloud costs?
The MCP servers themselves are open source and free (Apache-2.0). But they call your real AWS services, so creating resources or invoking paid APIs is billed by AWS as usual — read-only queries are usually cheap, but watch costs before creating or running resources.