Claude AWS API connectorWhat Claude can do in your AWS API account.
The Claude AWS API connector exposes 2 tools: 1 for reading, 1 that writes. 0 are undocumented. One suggests AWS CLI commands, the other runs them against your account with your local credentials. Below: the guardrails worth setting, and why AWS now points to a newer server.
Verified Trustpilot reviews · AI, automation & growth agency
What changes when Claude can run the AWS CLI for you
Instead of hunting through CLI reference pages for the right flags, you describe the task in plain words. This desktop extension lets Claude find the matching AWS CLI command and, if you let it, run that command on your account with the credentials already configured on your machine.
Get the right command first time. Ask how to list the buckets with public access and suggest_aws_commands returns the 5 most likely CLI commands with their full parameters, including commands released after Claude's training data stops.
Inspect your account in plain language. With call_aws, Claude executes the command and reads back the output, so you can ask which instances are running in a region without typing a line.
Change resources when your role allows it. The same tool can create, update and manage resources across AWS services. Whether it may do so depends on the IAM role behind your credentials, not on Claude.
What it will not do matters as much. A command on AWS's denylist never runs, and the server confirms changes only if you switch that option on. Nothing runs by itself either: no tool reacts to an alarm or a new resource. AWS itself now recommends its managed AWS MCP Server instead. Event-driven work belongs to automation platforms, covered from the Integrations hub.
Five terms worth knowing
The vocabulary around this extension, in one minute.
- Connector
- A link you set up once between Claude and an account you already have, so Claude can act on it while it answers you.
- Tool
- One named action the connector gives Claude. Claude picks the tools it needs by itself, and the directory lists them by name.
- Authorization
- The access you hand over once so Claude can work on a service. Here it rests on the AWS credentials configured on your machine.
- Approval
- The confirmation Claude waits for before an action that changes your account, displayed in the conversation when it applies.
- MCP
- The common standard connectors rely on: it is what lets an assistant like Claude send commands to an outside service such as AWS.
Set up AWS API MCP Server in three steps
- 01
Find the extension in Claude
In Claude Desktop, open the settings, go to Customize, then Connectors, and look up AWS API MCP Server. It is a desktop extension, installed on Claude Desktop. On Team or Enterprise, an Owner or Primary Owner enables it first.
- 02
Start the connection
Click Connect on its row. Access then comes from the AWS profile on your computer rather than from a web sign-in. If the extension stops responding, Disconnect it and connect it again.
- 03
Decide what the access covers
The service, not Claude, sets the reach of the access. Here that means the IAM role tied to your credentials: ReadOnlyAccess keeps Claude to non-mutating commands, AdministratorAccess opens creation and deletion.
The 2 tools: one suggests, one executes
AWS API MCP Server gives Claude 2 tools: 1 that read your account, 1 that change something in it.
A clean split: one tool proposes commands, the other runs them. Names stay exactly as Claude shows them.
- 1 read
- 1 write
What Claude reads (1)
1 toolOne tool that proposes commands and runs nothing.
suggest_aws_commands
Turns a plain-language request into the 5 most likely AWS CLI commands, each with a description and its full set of parameters. It proposes and stops there: nothing runs against your account.
What Claude can change (1)
1 toolOne tool that executes commands on your account, reads and writes alike. No source describes a default confirmation for it, so the rule further down applies.
call_aws
Approval: see the ruleExecutes an AWS CLI command against your account and returns the result, with validation and error handling along the way. Depending on the command, it lists resources or creates, updates and deletes them.
What gets checked before a command runs
By default, Claude stops and asks before each action it takes on an account for someone, right in the conversation.
On Team and Enterprise, owners decide whether members may let some actions through without a prompt each time, and they can cap what a connector may do for the whole organization. Claude works with the rights of the person connected and nothing beyond. For this extension, AWS adds its own layers: the IAM role behind your credentials is the primary control, a read-only mode and a consent mode can be switched on, and a security policy file can block or gate specific commands.
Which plans include it
None of the 819 sheets in the official directory states plan availability. That answer is published nowhere, connector by connector.
The general rule is public: remote connectors are open to all users on Claude, Cowork, Claude Desktop and mobile, while desktop extensions like this one install on Claude Desktop. On Team and Enterprise, an Owner or Primary Owner enables a connector before members use it. For the current state, check this extension's sheet in the official directory.
Where this extension stops
A connector is not an automation. Claude calls these tools while it answers you, so nothing fires on an alarm or a new deployment.
The partner badge in the directory is not a security audit, and Anthropic states on every sheet that it neither chooses a publisher's tools nor guarantees their behavior. Only AWS documents this extension; no Claude help page covers it. AWS's own setup guide recommends moving to the managed AWS MCP Server, which is a managed remote server and adds controls through IAM condition keys. The same directory rules apply on the Claude agility-cms connector page.
Need help connecting AWS API MCP Server to Claude?
A person reads every message.
