# AWS

> Inspect AWS resources, read CloudWatch logs, and audit IAM from an AI agent.

[Canonical HTML page](https://top-mcps.com/mcp/aws) · [server.json](https://top-mcps.com/mcp/aws.json) · [methodology](https://top-mcps.com/about/methodology)

## Install

### Claude Desktop — `claude_desktop_config.json`

Paste under mcpServers. Fully quit and reopen Claude after editing.

```json
{
  "mcpServers": {
    "aws": {
      "command": "uvx",
      "args": [
        "awslabs.aws-documentation-mcp-server"
      ],
      "env": {
        "AWS_PROFILE": "${AWS_PROFILE}",
        "AWS_REGION": "${AWS_REGION}"
      }
    }
  }
}
```

### Claude Code — `CLI or .mcp.json`

Run from your repo. Commit .mcp.json to share with your team.

```shell
# export AWS_PROFILE=agent-readonly
# export AWS_REGION=us-east-1
claude mcp add aws -- uvx awslabs.aws-documentation-mcp-server
```

### Cursor — `.cursor/mcp.json`

Global path: ~/.cursor/mcp.json. Reload window after editing.

```json
{
  "mcpServers": {
    "aws": {
      "command": "uvx",
      "args": [
        "awslabs.aws-documentation-mcp-server"
      ],
      "env": {
        "AWS_PROFILE": "${AWS_PROFILE}",
        "AWS_REGION": "${AWS_REGION}"
      }
    }
  }
}
```

### VS Code — `.vscode/mcp.json`

VS Code uses the "servers" key (not "mcpServers").

```jsonc
{
  "servers": {
    "aws": {
      "command": "uvx",
      "args": [
        "awslabs.aws-documentation-mcp-server"
      ],
      "env": {
        "AWS_PROFILE": "${AWS_PROFILE}",
        "AWS_REGION": "${AWS_REGION}"
      }
    }
  }
}
```

### Windsurf — `~/.codeium/windsurf/mcp_config.json`

Open via Cascade → hammer icon → Configure.

```json
{
  "mcpServers": {
    "aws": {
      "command": "uvx",
      "args": [
        "awslabs.aws-documentation-mcp-server"
      ],
      "env": {
        "AWS_PROFILE": "${AWS_PROFILE}",
        "AWS_REGION": "${AWS_REGION}"
      }
    }
  }
}
```

### Cline — `cline_mcp_settings.json`

Open via the Cline sidebar → MCP Servers → Edit.

```json
{
  "mcpServers": {
    "aws": {
      "command": "uvx",
      "args": [
        "awslabs.aws-documentation-mcp-server"
      ],
      "env": {
        "AWS_PROFILE": "${AWS_PROFILE}",
        "AWS_REGION": "${AWS_REGION}"
      }
    }
  }
}
```

### Continue — `~/.continue/config.json`

Continue uses modelContextProtocolServers with a transport block.

```json
{
  "experimental": {
    "modelContextProtocolServers": [
      {
        "transport": {
          "type": "stdio",
          "command": "uvx",
          "args": [
            "awslabs.aws-documentation-mcp-server"
          ],
          "env": {
            "AWS_PROFILE": "${AWS_PROFILE}",
            "AWS_REGION": "${AWS_REGION}"
          }
        }
      }
    ]
  }
}
```

### Codex CLI — `~/.codex/config.toml`

Codex uses TOML. Each server is a [mcp_servers.<name>] subtable.

```shell
# ~/.codex/config.toml
[mcp_servers.aws]
command = "uvx"
args = [
  "awslabs.aws-documentation-mcp-server",
]
env = { AWS_PROFILE = "${AWS_PROFILE}", AWS_REGION = "${AWS_REGION}" }
```

### Zed — `~/.config/zed/settings.json`

Zed calls them "context_servers". Settings live-reload on save.

```jsonc
{
  "context_servers": {
    "aws": {
      "command": {
        "path": "uvx",
        "args": [
          "awslabs.aws-documentation-mcp-server"
        ]
      },
      "env": {
        "AWS_PROFILE": "${AWS_PROFILE}",
        "AWS_REGION": "${AWS_REGION}"
      }
    }
  }
}
```

### ChatGPT — `ChatGPT → Apps directory`

AWS doesn't ship a hosted HTTPS endpoint today. ChatGPT supports remote MCP servers only — to use this server in ChatGPT you'll need to deploy it to a public HTTPS URL first (e.g. via Cloudflare Workers or Vercel) or wait for an official remote build.

```none

```

## At a glance

- **Maintainer:** AWS Labs
- **Transport:** stdio
- **Auth model:** API key
- **Required secrets:** AWS_PROFILE, AWS_REGION
- **Supported clients:** Claude, Cursor, VS Code, Windsurf, Any MCP-compatible client, AWS account
- **License:** Apache-2.0
- **Language:** Python
- **Latest version:** latest
- **Last verified:** 2026-06-02
- **GitHub stars:** 9,180 (fetched 2026-06-02T13:16:41.398Z)
- **Score:** 89/100 (rubric 2026-04 — see https://top-mcps.com/about/methodology)
- **Source:** https://github.com/awslabs/mcp

## Tools & permissions

| Tool | Description | Args | Side effects |
|------|-------------|------|--------------|
| `list_ec2_instances` | List EC2 instances in the configured region. | `—` | read |
| `list_s3_buckets` | List S3 buckets in the account. | `—` | read |
| `get_cloudwatch_logs` | Fetch log events from a CloudWatch log group. | `log_group: string, start_time?: string` | read |
| `list_iam_roles` | List IAM roles. | `—` | read |
| `get_caller_identity` | Return the IAM identity the MCP is acting as. | `—` | read |

## Security & scope

- **Access scope:** read-only
- **Sandbox:** Authenticates with the standard AWS credential chain — env vars, profile, IMDS, or SSO. Read-only at the tool layer; pair with an IAM role scoped to ReadOnlyAccess (or tighter) for defence in depth.
- **Gotchas:**
  - Never wire the agent to your developer admin profile — create a dedicated read-only IAM role.
  - CloudWatch Logs has per-request size limits; long log scans require paging.
  - STS GetCallerIdentity is exposed so the agent can verify which account it is acting against — check this before any cross-account work.

## Quick answer

**What it does.** Wraps the AWS SDK for read operations across EC2, S3, Lambda, RDS, IAM, CloudWatch, and STS. Exposes resource enumeration, log fetch, metric queries, and identity inspection to AI models.

**Best for:**
- Incident investigation
- IAM and security audits
- CloudWatch log search
- Cost-tagged resource enumeration
- Cross-service status checks

**Not for:**
- Provisioning new infrastructure
- Unattended writes to production
- Workloads needing AWS services outside the read surface

## Description

The official AWS MCP connects an AI agent to an AWS account via standard AWS credentials. Lists resources across the most-used services (EC2, S3, Lambda, RDS, IAM), reads CloudWatch logs and metrics, and runs read-only audits. The MCP is read-only by design — provisioning belongs in Terraform or CloudFormation, not in a conversation.

## Why it matters

AWS is the dominant cloud. An agent that can "tell me why my Lambda timed out last night" or "show me which S3 buckets are public" replaces a click-through tour of the AWS Console with a single prompt.

## Key features

- Read-only by design
- Standard AWS credential chain (env, profile, IMDS, SSO)
- CloudWatch logs + metrics
- IAM + STS audit tools
- Resource enumeration across EC2, S3, Lambda, RDS

## FAQ

### How does auth work?

The standard AWS credential chain — env vars, ~/.aws/credentials profile, IMDS on EC2, or SSO. Set AWS_PROFILE to point at a dedicated read-only profile rather than your admin role.

### Read-only?

Yes by design. The MCP intentionally does not expose Put/Create/Delete operations. For provisioning, drive Terraform or CloudFormation from the agent and review the diff before apply.

### Which services are covered?

The most-used: EC2, S3, Lambda, RDS, IAM, CloudWatch (logs + metrics), STS, plus a generic boto3 call tool for less-common services. The AWS Labs project is adding service coverage actively.

### How do I scope it down?

Use a dedicated IAM role with ReadOnlyAccess as the baseline and trim further. The MCP cannot enforce permissions the role itself does not have.

## Changelog

- **2026-06-02** — Refreshed install snippets and fact sheet; verified for 2026.
- **2025-04-15** — Initial directory listing.