aws-mcp
Runs on Cloudflare Workers to let MCP clients execute AWS API commands with your own credentials via SigV4-signed HTTPS requests. Supports common S3, EC2, IAM, Lambda, and STS operations plus a generic escape-hatch for any AWS API call.
README
aws-mcp
A remote MCP (Model Context Protocol) server that runs on Cloudflare Workers and lets an MCP client (like Claude) execute AWS commands with your own credentials.
How it works (and an important note about the "AWS CLI")
Cloudflare Workers run in a lightweight V8 isolate — there is no filesystem,
no subprocesses, and no Python, so the real aws CLI binary cannot run there.
Instead, this Worker talks to AWS the same way the CLI does under the hood: it
builds SigV4-signed HTTPS requests to the AWS service APIs. The result is the
same capability — "give it keys and run AWS commands" — implemented in a way that
actually runs on Cloudflare, with zero runtime dependencies (SigV4 is
hand-rolled with the Web Crypto API).
The MCP client (Claude) knows the AWS APIs, so it can drive the generic
aws_api_request tool to perform essentially any operation, plus there are
ready-made convenience tools for common tasks.
Tools exposed
| Tool | What it does |
|---|---|
aws_api_request |
Escape hatch — run ANY AWS API call (choose service, region, method, path, query, headers, body). Returns raw HTTP status + body. |
sts_get_caller_identity |
Who am I? Verifies the credentials are wired up. |
s3_list_buckets |
List all S3 buckets. |
s3_list_objects |
List objects in a bucket (optional prefix). |
ec2_describe_instances |
Describe EC2 instances in a region. |
ec2_describe_regions |
List available regions. |
iam_list_users |
List IAM users. |
lambda_list_functions |
List Lambda functions. |
Deploy to Cloudflare (via the GitHub form you have open)
The repo is already set up for Cloudflare's "Import a repository" flow:
- Build command: (leave empty)
- Deploy command:
npx wrangler deploy
Click Deploy. Cloudflare installs dependencies and runs Wrangler, which reads
wrangler.toml and publishes the Worker named aws-mcp.
After the first deploy — add your secrets
The Worker needs credentials. Set them as secrets (never commit them). Either in the dashboard (Workers & Pages → aws-mcp → Settings → Variables and Secrets → Add → Encrypt) or from a terminal:
npx wrangler secret put MCP_AUTH_TOKEN # a long random string YOU invent
npx wrangler secret put AWS_ACCESS_KEY_ID
npx wrangler secret put AWS_SECRET_ACCESS_KEY
# optional, only for temporary/STS credentials:
npx wrangler secret put AWS_SESSION_TOKEN
The default region lives in wrangler.toml (AWS_DEFAULT_REGION); override per
call with a region argument.
Security:
MCP_AUTH_TOKENis what stops the public internet from using your AWS keys. Always set it. Give AWS keys the least privilege you need — this server can do whatever those keys can do.
Connect an MCP client
Add the Worker URL (e.g. https://aws-mcp.<your-subdomain>.workers.dev) as a
Streamable HTTP MCP server, sending your token:
Authorization: Bearer <MCP_AUTH_TOKEN>
For Claude Code / CLI:
claude mcp add --transport http aws https://aws-mcp.<your-subdomain>.workers.dev \
--header "Authorization: Bearer <MCP_AUTH_TOKEN>"
Clients that can't set headers may instead append ?token=<MCP_AUTH_TOKEN> to the URL.
Quick checks
# health (no auth needed)
curl https://aws-mcp.<your-subdomain>.workers.dev/health
# list tools
curl -X POST https://aws-mcp.<your-subdomain>.workers.dev \
-H "Authorization: Bearer <MCP_AUTH_TOKEN>" \
-H "Content-Type: application/json" \
-d '{"jsonrpc":"2.0","id":1,"method":"tools/list"}'
# who am I
curl -X POST https://aws-mcp.<your-subdomain>.workers.dev \
-H "Authorization: Bearer <MCP_AUTH_TOKEN>" \
-H "Content-Type: application/json" \
-d '{"jsonrpc":"2.0","id":2,"method":"tools/call","params":{"name":"sts_get_caller_identity","arguments":{}}}'
Local development
npm install
cp .dev.vars.example .dev.vars # fill in your keys
npm run dev
Layout
src/index.ts MCP JSON-RPC over HTTP (Streamable HTTP transport) + routing/auth
src/aws.ts AWS request execution + tool definitions
src/sigv4.ts AWS Signature V4 signer (Web Crypto, no dependencies)
wrangler.toml Cloudflare Worker config
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。