---
title: "SageMaker AI MCP: review, setup & limitations · Undominated.ai"
canonical: https://undominated.ai/mcp-servers/awslabs-sagemaker-ai/
description: "AWS Labs MCP server for SageMaker HyperPod clusters on EKS or Slurm, intended for local stdio and read-only until write and sensitive-data flags are added."
---

# SageMaker AI MCP: review, setup & limitations · Undominated.ai

> AWS Labs MCP server for SageMaker HyperPod clusters on EKS or Slurm, intended for local stdio and read-only until write and sensitive-data flags are added.

[← Explore all mcp servers](/mcp-servers/)

CLOUD OPERATIONS / AWS Labs

# SageMaker AI MCP

AWS Labs MCP server for SageMaker HyperPod clusters on EKS or Slurm, intended for local stdio and read-only until write and sensitive-data flags are added.

 [See setup guidance ↓](#setup)[Original source ↗](https://github.com/awslabs/mcp/tree/8840003820a78dd018561d55c7ab12fcdd708464/src/sagemaker-ai-mcp-server)

SOURCE REVIEW

 Reviewed 2026-10-14
 Evidence 2 linked sources
 Publisher AWS Labs
 Licence [Apache-2.0 ↗](https://github.com/awslabs/mcp/blob/8840003820a78dd018561d55c7ab12fcdd708464/src/sagemaker-ai-mcp-server/LICENSE)
 Revision 8840003820a7
 [Read what was—and wasn’t—checked ↓](#review)

“Manages HyperPod deployment, nodes, and lifecycle, and says not to modify the server for network use.”

 [AWS Labs · upstream description ↗](https://github.com/awslabs/mcp/blob/8840003820a78dd018561d55c7ab12fcdd708464/src/sagemaker-ai-mcp-server/README.md) Our analysis follows below.

01 / THE REASONING

## Why this made the selection.

 - The catalogue has no SageMaker or HyperPod MCP server.
- The README separates read-only tools from the write tools that need --allow-write.

### A good fit for

 - Inspecting a HyperPod cluster from a local agent
- Deploying or deleting a HyperPod stack only after --allow-write is set

### Weigh up before choosing

 - This review did not execute it.
- This review did not start the server.
- The example buttons and JSON enable --allow-write and --allow-sensitive-data-access.
- Tool names were taken from the README's mode list, not from a tools/list call.

02 / THE REVIEW RECORD

## What we actually inspected.

Source review has boundaries. A clear record is more useful than a “safe” badge.

### Material inspected

 - src/sagemaker-ai-mcp-server/README.md
- src/sagemaker-ai-mcp-server/LICENSE

### Our findings

 - The current surface is SageMaker HyperPod on Amazon EKS or Slurm, including deployment, node management, and lifecycle operations.
- The README says the server is read-only by default and names manage_hyperpod_stacks deploy and delete, plus node update_software and batch_delete, as write operations that need --allow-write.
- It says the server is for STDIO mode only, runs with the user's permissions and complete filesystem access, and should not be modified for network use. Its directory LICENSE is Apache-2.0.

### Not established by this review

 - The server was not started.
- No HyperPod tool was called.
- No cluster was changed.

The review applies to the material and revision named here. A newer upstream release can change its behavior.

03 / PUT IT TO WORK

## Connect a server deliberately.

[Upstream setup instructions ↗](https://github.com/awslabs/mcp/blob/8840003820a78dd018561d55c7ab12fcdd708464/src/sagemaker-ai-mcp-server/README.md)
 - Client examples use uvx awslabs.sagemaker-ai-mcp-server@latest inside JSON, not a single shell line.
- Drop --allow-write to stay in the read-only mode the README recommends.
- The directory LICENSE was fetched separately and is Apache-2.0.

### Before you start

 - uv
- AWS credentials with access to the HyperPod cluster
- A local stdio MCP client

### Compatibility

Kiro · Cursor · VS Code

### Transports

stdio

### Local stdio, read-only unless flagged

 - Read HyperPod cluster state
- Deploy or delete stacks and update nodes only with --allow-write
- Read logs and events when --allow-sensitive-data-access is set

### Cost model

Apache-2.0; AWS credentials are required and were not used

04 / FOLLOW THE EVIDENCE

## The source trail.

Our notes are separate from the original resource. Check upstream before adopting a new version.

 - [SageMaker AI README ↗](https://github.com/awslabs/mcp/blob/8840003820a78dd018561d55c7ab12fcdd708464/src/sagemaker-ai-mcp-server/README.md) Checked 2026-10-14 https://github.com/awslabs/mcp/blob/8840003820a78dd018561d55c7ab12fcdd708464/src/sagemaker-ai-mcp-server/README.md Supports: STDIO warning and write flag
- [Directory LICENSE ↗](https://github.com/awslabs/mcp/blob/8840003820a78dd018561d55c7ab12fcdd708464/src/sagemaker-ai-mcp-server/LICENSE) Checked 2026-10-14 https://github.com/awslabs/mcp/blob/8840003820a78dd018561d55c7ab12fcdd708464/src/sagemaker-ai-mcp-server/LICENSE Supports: Apache-2.0

KEEP COMPARING

## Other approaches to consider.

Related by category or shared topics. These are alternatives to inspect, not a measured quality order.

 [### AWS EKS MCP Server ↗ Inspects Amazon EKS. It is not a generic Kubernetes server.](/mcp-servers/awslabs-eks/)[### AWS Billing and Cost Management MCP Server ↗ Reads realised AWS billing and cost-management data, which is a different number from the advertised unit price.](/mcp-servers/awslabs-aws-billing-cost-management/)[### AWS CloudTrail MCP Server ↗ Queries CloudTrail events for an account you administer.](/mcp-servers/awslabs-cloudtrail/)

 [AI Tools ↗](/tools/)[Skills ↗](/skills/)[Agents ↗](/agents/)[MCP Servers ↗](/mcp-servers/)[Workflows ↗](/workflows/)

## Continue your investigation

 - [Inspect resource evidence](/tools/)
- [Choose a host](/agents/)
- [Inspect reusable workflows](/skills/)
- [Use Undominated data](/api/)
