AWS Labs publishes an open-source suite of MCP servers rather than one endpoint — separate servers for documentation, pricing, CDK, Cloud Control API, and more. You install the ones you need. This page covers what each does and which to reach for first.
Bring your own
Claude Sonnet 4.5
Chat with 60+ AI models on the same workflow — switch to a different model mid-conversation and re-run the same prompt, or use Compare mode to put several side-by-side and balance quality vs. cost.
Uses your standard AWS credential chain — profile, environment variables or SSO. The agent inherits that profile's IAM permissions exactly.
How models use it and what it is built for.
The awslabs/mcp repository hosts a family of focused MCP servers, each wrapping a different slice of AWS. The Documentation server searches and retrieves AWS docs so an agent stops guessing at service limits and parameter names. The Cloud Control API server creates, reads, updates, deletes and lists AWS resources from natural language, covering hundreds of resource types through one uniform interface. Others cover pricing, CDK, Terraform, diagrams and per-service APIs. They run locally over stdio and authenticate with your existing AWS credential chain, which means an agent inherits exactly the IAM permissions of the profile you point it at.
Typical tools an AI model can call. Exact names vary by version.
Copy any of these into MCP Agent Studio after connecting.
What are the current limits on Lambda concurrent executions, and how do I raise them?
List every S3 bucket in us-east-1 and flag any without default encryption.
Estimate the monthly cost of an RDS db.r6g.large running 24/7 in eu-west-1.
Which CDK construct should I use for an ALB in front of a Fargate service?
This is not a single-model product: you get the same MCP connection with 60+ models (Claude, GPT, Gemini, DeepSeek, open-weight, and more), you can switch mid-conversation, and you can open Compare mode to run the same prompt against multiple models at once. The card above is a suggested starting point for this server — not the only choice.
Default pick for AWS
Claude Sonnet 4.5
AWS work is multi-step and unforgiving about parameter names. Sonnet 4.5 chains documentation lookups into concrete actions without inventing service limits.
Listing tools proves the server is reachable, not that a model can work with it. Evals go further: they read every tool on the server, write a test suite from its real schemas, and run it — code decides pass/fail on the responses (schema conformance, error codes, pagination, result caps) while a scoring model grades plain-English tasks driven through the tools.
Get a pass/fail report per tool with the evidence behind each verdict — and replay the same suite after every schema change. Destructive tools are excluded from the run.
Run evalsOpen MCP Agent Studio with the connection pre-filled. Add your token, pick any of 60+ models, and start chatting — no install required.
Open Agent StudioCommon questions about connecting, scoping and using it safely.
No, and that trips people up. AWS Labs ships a suite of focused servers in the awslabs/mcp repository — documentation, pricing, CDK, Cloud Control API, networking and more. You install the ones matching your work rather than connecting to one endpoint.
The documentation server. It is read-only, so there is no blast radius, and it fixes the most common failure — an agent inventing service limits or parameter names. Add the Cloud Control API server once you actually want the agent changing resources.
They use the standard AWS credential chain: a named profile, environment variables or SSO. That means the agent has exactly the IAM permissions of whichever profile you point it at, so scoping that profile tightly is your main safety control.
Only with a scoped IAM role. Start read-only, and if you enable the Cloud Control API server, use a profile limited to a non-production account. The agent will do exactly what its credentials permit, which is the whole risk.
Local. They run over stdio and are distributed through PyPI and container images rather than as a hosted endpoint, which is why they use your local credential chain instead of asking for a key.