sagemaker-ai

This MCP server provides a comprehensive interface to manage and operate Amazon SageMaker AI resources through a suite of tools. It exposes endpoints for SageMaker AI endpoints and endpoint configurations, a broad set of training, processing, transform, and inference-related jobs, pipelines, domains, models, model cards, apps, and an MLflow tracking server, all via the MCP tooling layer. Using the listed tools, you can list, describe, create, delete, and stop resources, enabling you to automate common SageMaker AI workflows from a single, consistent MCP interface. The server authenticates to AWS using the AWS profile specified in AWS_PROFILE and operates against the region defined by AWS_REGION, with SageMaker execution permissions granted by SAGEMAKER_EXECUTION_ROLE_ARN.

Prerequisites:\n- Python 3.10 installed in your environment.\n- uv (Astral uv) installed and accessible.\n- AWS credentials configured (via AWS_PROFILE or environment variables) with permissions to access SageMaker AI resources.\n\nInstallation steps:\n1) Install uv (Astral uv) and Python 3.10: Follow the instructions from the uv documentation to install uv and then install Python 3.10 via uv if you’re using uv to manage runtimes. Example (conceptual):\nbash\n# Install uv (per Astral docs)\n# Install Python 3.10 via uv (as suggested in the README)\nuv python install 3.10\n\n\n2) Ensure AWS credentials are configured: create or update your AWS credentials file or set environment variables. For example:\nbash\nexport AWS_PROFILE=your-aws-profile\nexport AWS_REGION=us-east-1\nexport SAGEMAKER_EXECUTION_ROLE_ARN=arn:aws:iam::123456789012:role/SageMakerExecutionRole\n\n\n3) Run the MCP server via uvx (no separate installation required for the server package):\nbash\nuvx dgallitelli-sagemaker-ai-mcp-server\n\n\nNote: If you publish this MCP server as a package, you can also install it with your preferred package manager and then run it with uvx or the configured runtime, depending on how you package it.

Tips and caveats:\n- Ensure that the AWS profile used has explicit permissions for all SageMaker AI resources you intend to manage (endpoints, endpoint configs, training/processing/transform jobs, pipelines, domains, models, model cards, apps, and MLflow tracking servers).\n- The server relies on AWS credentials; keep them secure and rotate them following your organization’s policy.\n- By default, AWS_REGION defaults to us-east-1 if not provided; override this with AWS_REGION to target your region.\n- If you encounter permissions-related errors, verify that the SageMaker execution role ARN in SAGEMAKER_EXECUTION_ROLE_ARN has the necessary trust and permissions.\n- When using MLflow or other SageMaker managed tracking services, ensure the required networking and IAM permissions are in place to allow API calls and data transfer.\n- If you update or add resources, you may need to refresh your AWS credentials or re-authenticate your AWS profile.\n- The MCP server exposes a large number of tools; consider grouping and scripting common workflows to minimize manual command entry.\n- If running behind a proxy, ensure the MCP server process can access AWS endpoints through the proxy configuration.