AWS

by awslabs

Community Cloud Infrastructure 10k likes

A collection of 60-plus purpose-built MCP servers from AWS Labs, each scoped to one service or workflow — documentation lookup, CDK/CloudFormation, DynamoDB, Bedrock, cost management, and more.

Add to Claude Code (example: docs server)

claude mcp add aws-documentation -- uvx awslabs.aws-documentation-mcp-server@latest

Source: https://github.com/awslabs/mcp

About

Rather than one generic "AWS" MCP server, awslabs/mcp ships dozens of narrow, single-purpose servers you install individually depending on what you're building: an AWS Documentation server for searching and reading official docs, a CDK/CloudFormation/Terraform-adjacent Infrastructure-as-Code server, database servers for DynamoDB, Aurora, DocumentDB, and Redshift, ML-focused servers for Bedrock and SageMaker, plus operational servers for CloudWatch, IAM, and cost/pricing analysis. Each is independently versioned and published to PyPI.

The point of splitting them up is context economy and safety scoping: an agent working on a Lambda deployment doesn't need — and shouldn't have — tools for touching Aurora or IAM. You wire up only the servers relevant to the task, and each one encodes AWS-specific best practices (correct CDK constructs, current API shapes, valid resource configurations) so the agent's output matches what AWS actually recommends rather than what's in its training data.

Key features

  • 60-plus independently installable MCP servers, one per AWS service or workflow area
  • AWS Documentation server for search and up-to-date doc retrieval, reducing hallucinated API details
  • Infrastructure-as-Code servers for CDK, CloudFormation, EKS, ECS, and container tooling
  • Database-specific servers for DynamoDB, Aurora (Postgres/MySQL), DocumentDB, Neptune, and Redshift
  • ML/AI servers for Bedrock, SageMaker, and Kendra
  • Cost, pricing, and CloudWatch operational servers for monitoring and spend analysis

Use cases

  • Looking up current AWS documentation mid-task instead of relying on a model's outdated training data
  • Having an agent scaffold a CDK stack that follows AWS's own recommended patterns
  • Querying DynamoDB or Aurora schemas and data directly from a coding agent
  • Pulling cost and pricing data into an agent workflow before recommending an architecture change

Available tools

aws-documentation-mcp-server

Searches AWS documentation and returns current, accurate page content and recommendations.

aws-api-mcp-server

Lets an agent run AWS CLI-equivalent operations against your account through MCP.

cdk-mcp-server

Guides and validates AWS CDK infrastructure code against current best practices.

dynamodb-mcp-server

Reads and manages DynamoDB tables and items directly from an MCP client.

cost-analysis-mcp-server

Surfaces AWS cost and usage data for a workload before or after deployment.

Frequently asked questions

Do I install one 'AWS MCP server' or many?

Many — it's a suite of independent servers; you pick and install only the ones relevant to your task, such as the documentation server or the DynamoDB server.

Does an agent need my AWS credentials to use these?

For servers that touch real resources (DynamoDB, CDK deploys, cost data), yes — they use your existing AWS credentials/profile the same way the AWS CLI would; documentation-only servers need no AWS access at all.