AWS open source MCP servers (mcp) is a collection of dedicated MCP servers maintained by awslabs on GitHub, designed to enable AI applications to make fuller use of AWS in any scenario where MCP is used. MCP in the project name refers to Model Context Protocol, an open protocol for seamless integration between LLM applications and external data sources and tools.
Purpose and Background An MCP server is a lightweight program that exposes specific capabilities through the standardized MCP protocol. Host applications (such as chatbots, IDEs, and other AI tools) have built-in MCP clients and maintain one-to-one connections with MCP servers. Common MCP clients include agentic AI coding assistants (such as Kiro, Cline, Cursor, Windsurf) and chat applications such as Claude Desktop. MCP servers can access local data sources and remote services to provide additional context for models, thereby improving generated results.
MCP Servers for AWS provide AI applications with access to AWS documentation, contextual guidance, and best practices through this protocol. The repository explains that MCP servers can enhance the capabilities of foundation models in several ways:
- Improve output quality: placing relevant information directly into the model context can reduce hallucinations, provide more accurate technical details, generate more precise code, and make recommendations align with current AWS best practices and service capabilities.
- Access the latest documentation: foundation models may not know about recently released APIs or SDKs, and MCP servers can pull the latest documentation to ensure AI assistants always work based on the latest AWS capabilities.
- Workflow automation: turn common workflows into tools that foundation models can use directly, covering AWS-related processes such as CDK and Terraform, enabling AI assistants to complete complex tasks more accurately and efficiently.
- Specialized domain knowledge: provide in-depth contextual knowledge of AWS services that may not be fully covered in foundation model training data, helping cloud development tasks obtain more accurate and useful responses.
Available Servers and Organization The repository organizes MCP servers along two dimensions: "what you are building" and "how you work." By what is being built, they can be divided into: infrastructure and deployment (container platforms, serverless and function support), AI and machine learning, data and analytics (SQL and NoSQL databases, search and analytics, backend API provsuppliers, caching and performance), developer tools and support, integration and messaging, cost and operations, healthcare and life sciences, etc. By working style, they can be divided into: Vibe Coding and development (core development workflows, infrastructure as code, application development, container and serverless development, testing and data, life sciences workflow development, healthcare data management), conversational assistants (knowledge and search, content processing and generation, business services), autonomous background agents (data operations and ETL, caching and performance, workflows and integration, operations and monitoring). The repository also includes the MCP AWS Lambda Handler module, a Python library for creating serverless HTTP handlers using AWS Lambda, with capabilities such as pluggable session management (built-in DynamoDB support), customizable authentication and authorization, and more.
Usage and Installation The MCP servers in this repository are designed to support only stdio transport. As of May 26, 2025, all MCP servers' latest major versions have removed Server Sent Events (SSE) support to comply with the MCP specification's backward compatibility guidelines; applications that still need SSE should use the previous major version of the corresponding server. The repository is advancing support for Streamable HTTP. The documentation provides installation and setup instructions for macOS/Linux and Windows, methods for running MCP servers in containers, and getting-started configurations for various clients, such as Kiro (~/.kiro/settings/mcp.json), Cline and Amazon Bedrock (cline_mcp_settings.json), Cursor (.cursor/mcp.json), Windsurf (~/.codeium/windsurf/mcp_config.json), VS Code (.vscode/mcp.json), Claude Code (.mcp.json), fx (~/.fx/mcp.json). The repository also provides one-click install buttons to install servers directly in Cursor or VS Code, along with Vibe coding examples and the Vibe Coding with AWS MCP Servers workshop.
Choosing Between Local and Remote The repository explains the applicable scenarios for local and remote MCP servers. Local serversare suitable for development and testing, offline work, data privacy (sensitive data and credentials remain local), low latency, resource control, etc. Remote servers are suitable for team collaboration, resource-intensive tasks, always available, automatic updates, scalability, security (centralized IAM permission control, zero credential exposure), and governance (audit logs and compliance monitoring).
Applicable Scenarios Suitable for scenarios such as using AI coding assistants for cloud-native development, infrastructure as code, container and serverless development, data and analytics, integration and messaging, cost and operations, healthcare and life sciences workflows, as well as building customer-facing chatbots, autonomous background agents, and more.
Limitations and Considerations The repository explicitly reminds users that they must ensure their use of these servers complies with their applicable terms and any applicable laws, rules, regulations, policies, or standards. The repository still works and accepts contributions, but AWS Labs has launched Agent Toolkit for AWS as its successor, recommended for production software or scenarios building agents for customers; over time, the most useful projects in this repository will migrate to Agent Toolkit for AWS. Regarding specific prices, fee standards, performance metrics, etc., the official text does not currently provide them.
