Labsco
MCP SERVER · OFFICIAL PROJECT

AWS MCP Servers

by awslabs ↗

AWS Labs' suite of MCP servers — documentation, API access, infrastructure as code, databases, cost and more — each installed separately with uvx.

MCP Plumbing: Proxies, Registries & Inspectors✓ Official source
Summary
Not a server — a catalogue you install from, one service at a time.

Treat the repository as an index: pick the two or three servers matching the work in front of you, because each one adds its own tools and its own credential requirement to the client. The documentation and pricing servers are the ones that pay off fastest, since they close the gap between what the model remembers and what AWS shipped this quarter.

What it is

The AWS Labs monorepo of open-source MCP servers for AWS. It is not one server: each subdirectory publishes its own package with its own tools, credentials and configuration, and you install only the ones a project needs. They span AWS documentation and API access, infrastructure as code, containers and serverless, databases and analytics, AI and Bedrock, messaging, cost and operations, and healthcare and life sciences. All of them speak stdio.

What you get
  • Current AWS documentation and API references in the model's context, so answers track releases rather than training data — `awslabs.aws-documentation-mcp-server`
  • A remote, AWS-hosted knowledge server covering docs, What's New posts, Builder Center, blog posts and Well-Architected guidance — `aws-knowledge-mcp-server`
  • AWS API operations reachable from the client, alongside a core server that routes to the rest — `awslabs.aws-api-mcp-server`, `awslabs.core-mcp-server`
  • Infrastructure-as-code workflows for CloudFormation and Terraform, and support case handling — `awslabs.aws-iac-mcp-server`, `awslabs.aws-support-mcp-server`
  • Databases and analytics: Aurora DSQL, MySQL, Neptune, Keyspaces, Valkey caching and the data-processing server — `awslabs.aurora-dsql-mcp-server`, `awslabs.mysql-mcp-server`, `awslabs.valkey-mcp-server`, `awslabs.aws-dataprocessing-mcp-server`
  • Bedrock knowledge-base retrieval and other AI/ML servers for agent-side workflows — `awslabs.bedrock-kb-retrieval-mcp-server`
  • Cost work, including pricing lookups for estimating a design before it is built — `awslabs.aws-pricing-mcp-server`
  • A container image per server in the public AWS ECR gallery, for clients that would rather not run uvx
  • One-click install links for Kiro, Cursor and VS Code on each server's row in the repository README
Requirements

AWS credentials with access to whichever services the chosen server touches — the permissions question is per server, not per repository. Install `uv`, then `uv python install 3.10`, and add the server to the client as `uvx awslabs.<name>@latest`; Windows uses the `uv tool run --from` form instead. `FASTMCP_LOG_LEVEL` is set in the env block of every documented config, and some servers take their own, such as `AWS_DOCUMENTATION_PARTITION`. Keeping the `@latest` suffix re-checks PyPI on every client start, which costs load time — drop it and manage the uv cache yourself if startup matters. These servers support stdio only: Server-Sent Events support was removed from all of them on May 26th, 2025. AWS now positions the Agent Toolkit for AWS as the successor to these servers and expects the most useful projects here to move into it over time; the repository continues to work and accept contributions. Apache-2.0.

Setup effort

One command plus a key — uvx awslabs.core-mcp-server@latest, then supply credentials