CloudSword: cloud security testing
CloudSword
A security testing tool for Alibaba, Tencent and other Chinese cloud accounts that also runs as an MCP server for an agent.
- 626 ★ on the repository
- Last commit
- Apache-2.0
Broad access
Why this level
- It needs access keys of a cloud account, and in MCP mode module parameters arrive as call arguments and can end up in the conversation with the model
- Some modules change the account: they create RAM and CAM users, attach policies, issue access keys and deploy a cloud function
- It is a dual-use tool: suitable for authorized assessments and equally for reconnaissance of accounts you do not own
Install
- Works in
Manual install
brew tap wgpsec/tap
brew install wgpsec/tap/cloudswordInstalls with Homebrew. Prebuilt binaries for other systems are on the repository's releases page.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- It needs access keys of a cloud account, and in MCP mode module parameters arrive as call arguments and can end up in the conversation with the model
- Some modules change the account: they create RAM and CAM users, attach policies, issue access keys and deploy a cloud function
- It is a dual-use tool: suitable for authorized assessments and equally for reconnaissance of accounts you do not own
This is third-party code. Review the repository files before installing.
What it does
CloudSword, also known as 云鉴, is a Go tool for security specialists. It lists the resources in a cloud account, looks for weak spots and can switch on simple protective measures. It supports Alibaba Cloud, Tencent Cloud, Huawei Cloud, Baidu Cloud and Qiniu. The console works like Metasploit, with use, set and run commands, module search, per-module help and input hints. Access keys stay in memory and are not written to disk. There are few modules, and they vary in depth. Alibaba Cloud has an inventory of OSS, ECS, RAM and domains, object search in OSS buckets, a rule that limits a bucket to image uploads, and RAM user management: listing, creating, attaching policies, access keys. Tencent Cloud has an inventory of COS, CVM, Lighthouse and CAM, the same user operations for CAM and the creation of a decoy access key. Huawei Cloud, Baidu Cloud and Qiniu have one module each: list buckets. MCP mode arrived in version 0.0.2. The command `cloudsword sse http://localhost:8080` listens on port 8080 over SSE, and any other first argument starts a stdio server. Each module becomes its own MCP tool, and module parameters arrive as call arguments. The agent picks the module, runs it and returns the console output.
Who it is for. For security specialists who own accounts on Chinese clouds and want an agent to run the checks.
Good fit when
- You need an inventory of your own Alibaba Cloud or Tencent Cloud account: OSS and COS buckets, ECS and CVM servers, RAM and CAM users
- An authorized cloud assessment is under way and the agent should run modules in turn and collect the output into a report
- You want to see which users, policies and access keys exist in an account before changing anything
- You want to plant a decoy access key in Tencent Cloud to notice anyone trying leaked credentials
- An OSS bucket should accept images only, and a ready module is the quickest way to apply the limit
Not a fit when
- Accounts you do not own and lack written permission to test: the tool can create users and issue keys
- AWS, Azure, Google Cloud and Russian clouds: the module list has none for them
- Regular compliance reports: CloudSword runs single modules on request and builds no reports
- An English console and documentation: the tool's output is Chinese, and so is its manual
- Keys with broad permissions should not be typed into a chat with a model: in MCP mode they arrive as call arguments
Example request
List the COS buckets, CVM servers and CAM users in my Tencent Cloud account and put them into one table.MCP
- Transport
- stdio, sse
- Authentication
- API key
Environment variables
CLOUD_SWORD_ACCESS_KEY_IDrequired- Access key ID of the cloud account.
CLOUD_SWORD_ACCESS_KEY_SECRETrequired, secret- Secret part of the access key.
CLOUD_SWORD_SECURITY_TOKENsecret- Security token for temporary credentials.
CLOUD_SWORD_DETAIL- Detailed output: yes or no.
Limitations
The console, module tables and manual are written in Chinese, with no Russian or English version. Running modules needs access keys of a cloud account, plus a security token when the credentials are temporary. The module list is short: Huawei Cloud, Baidu Cloud and Qiniu only list buckets, and AWS, Azure and Google Cloud are not supported at all.
In SSE mode the port is taken from the address you pass, so give the full address with scheme and port, for example http://localhost:8080. The repository was last updated on February 3, 2026, so check the current state of the cloud providers' APIs yourself.
Whether the Chinese cloud consoles are reachable from Russia was not checked for this card.
How to disable. Remove the server from the client configuration, in Claude Code that is `claude mcp remove cloudsword`, and uninstall the program with `brew uninstall cloudsword`. CloudSword does not write your access keys to disk, but the keys, RAM and CAM users and cloud functions it created must be revoked in the provider's console.
FAQ
How do I connect CloudSword to Claude Code?
Install the program with Homebrew and add the server with `claude mcp add cloudsword -- cloudsword stdio`. In the code, any first argument other than sse starts stdio mode. For SSE clients, run `cloudsword sse http://localhost:8080` and point the client at http://localhost:8080/sse.
Which clouds does CloudSword support?
Alibaba Cloud, Tencent Cloud, Huawei Cloud, Baidu Cloud and Qiniu. Alibaba Cloud and Tencent Cloud have the most modules: resource inventory, users, access keys. The others only list buckets.
Which CloudSword modules change the state of an account?
Creating RAM and CAM users, attaching policies to them, setting up console login and issuing access keys, plus limiting an OSS bucket through a cloud function and creating a decoy key for Tencent Cloud. The remaining modules only read lists.
What does the create_honey_token module do?
Module 2102 for Tencent Cloud creates a decoy, an access key that no real user relies on. Any attempt to use it signals a data leak.
How does CloudSword differ from Prowler?
Prowler for Claude Code brings an account in line with CIS, PCI-DSS and other standards. CloudSword works with Chinese clouds, runs single modules and can create objects in the account.
Discussion
No questions or reviews yet.
Related
MCP serversMCP-Shield finds your MCP configs, connects to servers and scans tool descriptions for hidden commands, exfiltration channels and tool shadowing
Open-source deception framework: decoy SSH, HTTP, database and MCP services in YAML, optional LLM responses, attacker activity logged as evidence
An offensive security MCP server: the agent runs 150+ pentest and recon tools and drives automated target assessments
An MCP server that gives the agent vulnerability triage: NVD data, EPSS, the CISA KEV catalog, MITRE ATT&CK mapping and one-call CVE risk scoring
An authorized bug-bounty suite: 50 subagents, 26 commands and 2 bug-bounty-platform MCP servers for Claude Code, Codex and five more agents
Security scanner for agent skills and MCP servers: finds prompt injection, data exfiltration and supply chain risks before install
README in short
The Chinese-language document opens with the purpose: quickly show a cloud tenant its resources and weak spots and suggest protective measures. Four strengths follow: Chinese output, command hints, Metasploit-style logic and keys that never touch disk. Installation goes through Homebrew or a binary from the releases page, and a detailed manual lives at wiki.teamssix.com. After that come a section on MCP with connection screenshots, module tables per provider with a usefulness rating, and an example run of the module that lists OSS buckets.