Connect your AI coding agent to AWS
Connect your coding agent to current AWS documentation, skills for commons tasks, and your AWS account with the Agent Toolkit for AWS.
If you are building apps in 2026, you are probably using an AI coding agent to help you write code. I use coding agents too, and it's a great way to get started quickly and ship things faster. What does get frustrating is when the agent recommends something that doesn't work as expected and you need to go figure out what's wrong.
One reason this happens is that every model has a training cutoff date, while AWS services (and technology in general) keep changing. If a feature launched or an API changed after that date, your agent may still recommend the older approach without even knowing that the information is incorrect.
What the Agent Toolkit for AWS is
The Agent Toolkit for AWS gives your coding agent access to AWS docs, instructions for common tasks, and AWS APIs. Instead of relying only on what the model learned during training, your agent can check AWS sources while it works.
It does that through three parts:
- AWS documentation and API references. Your agent checks the documentation at the moment you ask, rather than relying only on what it remembers from training.
- Skills. Task-specific guides for common AWS workflows that your agent can use.
- Access to your AWS account. When you ask it to do something for you, your agent can call AWS APIs within your existing permissions.
It works with the agent you already use: Kiro, Claude Code, Codex, Cursor, and others. So you don't need to learn a new tool.
How it works
The Agent Toolkit combines two pieces: the AWS MCP Server and AWS skills.
The AWS MCP Server gives your coding agent a set of tools it can call through MCP (Model Context Protocol) . Those tools let it search AWS documentation, retrieve service information, and make AWS API calls.
Skills provide task-specific instructions written by the AWS teams that build those services. A skill guides the agent through a workflow, including what to check and which MCP tools to use. They're open source too.
What you will need
- Your favorite AI coding agent
- An AWS account
Step 1: Choose one setup path
There are two ways to set up the Agent Toolkit. Your just need to pick and run one of them.
The easiest path: let your coding agent do it
Sign in to the AWS Console . At the top of Console Home, you'll see the Agent Toolkit for AWS banner. Click Get setup prompt, then paste the prompt into your coding agent's chat. If you use an agent in your terminal, paste it into the agent, not into the regular shell.
Your agent follows the public setup instructions : it installs the AWS CLI if you don't have it, signs you in with
aws login, connects to the AWS MCP Server, and installs the core AWS skills. You answer a few questions along the way, but the agent handles the setup.The manual path: run the setup yourself
If you're like me and want to understand what the prompt changes before you run it, we can configure the AWS CLI, server connection, and skills ourselves.
If you haven't installed the AWS CLI or used
aws login before, this development environment guide walks through both.First, make sure you have the AWS CLI , version 2.35.0 or later:
1
aws --versionThen sign in to your AWS account:
1
aws loginNow run the Agent Toolkit setup wizard:
1
aws configure agent-toolkitThe wizard finds the coding agents installed on your machine, connects them to the AWS MCP Server, and installs the default AWS skills.
Restart your agent so it picks up the new configuration.
Step 2: Make sure it actually worked
Start a new conversation with your agent and ask something that only works if the server is connected:
What AWS Regions are available?
If you get a list of regions back, you're connected.
What you can ask for
You don't need to know what skills exist or which service you need. Ask the question you actually have:
I want to build an app where people upload photos and see them in a gallery. What should I use on AWS?
Add a database to this app so users can save their favorites.
My deployment failed and I can't tell why. Can you take a look?
Your agent figures out which skills apply and reads the current docs before answering.
When you're ready to build in your account
Both setup paths sign you in with
aws login before configuring the Agent Toolkit. The AWS MCP Server then uses the permissions of that signed-in AWS identity for both read-only and write API calls.Your agent can only do what your own AWS user is allowed to do. If you want it more restricted than you are, AWS lets you write permissions that apply specifically to agent requests, so your agent stays read-only somewhere your own account can write: Understanding IAM for Managed AWS MCP Servers .
Adding more skills
Your agent discovers skills on its own, so most of the time you can ignore this. When you want to browse the catalog yourself, the AWS CLI (version 2.35.0 or later) lists it for you:
1
2
3
aws agent-toolkit list-available-skills
aws agent-toolkit search-skills --search-query serverless
aws agent-toolkit add-skill --skill-name aws-serverlessSkills get revised as services change, so
aws agent-toolkit update-skill is worth running now and then. They install globally, not per project, so you do it once and every project gets them.If something doesn't work
If you were already using
aws-api-mcp-server or aws-knowledge-mcp-server, remove them. The Agent Toolkit replaces both, and running them side by side leaves your agent choosing between overlapping tools.If you hit an authentication error, check the troubleshooting guide .
What this costs
The Agent Toolkit is free. You only pay for the AWS resources your agent actually creates, at normal prices, and plenty of services have a Free Tier .
What you just set up
- AWS MCP Server: your agent reads current AWS docs instead of working from memory
- AWS Skills: instructions for common AWS tasks, loaded when your question calls for them
Ask your agent for something you'd normally look up yourself, and see what it comes back with. Let me know in the comments if you have any questions or run into any trouble with the setup!
Enjoyed reading this content? Let the author know!
Your likes, comments, shares, and saves help creators reach more builders.
Loading recommendations
Loading article