MCP.so
登录

Cloud Ai Wallet

@browtastic

关于 Cloud Ai Wallet

MCP server for AI agent research — captures LLM reasoning, model identity, and feedback via schema injection

基本信息

分类

AI 与智能体

许可证

MIT

运行时

python

传输方式

stdio

发布者

browtastic

提交者

Hussain Mahmood

配置

使用下面的配置,将此服务器添加到你的 MCP 客户端。

{
  "mcpServers": {
    "aicryptovault": {
      "type": "sse",
      "url": "https://agent.cloudaiwallet.com/sse"
    }
  }
}

工具

未检测到工具

工具是从 README 中自动提取的。维护者可以在 ## Tools 标题下列出工具,即可填充这部分内容。

概览

What is Cloud Ai Wallet?

Cloud Ai Wallet is a deliberately vulnerable MCP honeypot server for defensive AI security research. It exposes a naive LLM assistant ("Vela") via a chat UI and MCP bait tool-servers, allowing researchers to study how attackers exploit agentic AI and the Model Context Protocol. All attacker interactions are captured for analysis.

How to use Cloud Ai Wallet?

Set up the server in a disposable, isolated environment. Replace all placeholder config values (e.g., <GATEWAY_TOKEN>, <NEO4J_PASSWORD>, acv-canary-*-PLACEHOLDER) with your own bait secrets. Run the components as described in the repository (nova-chat, mcp-servers, honeypot-skillwatch, etc.). The host RCE via the devtools MCP server is intentional bait.

Key features of Cloud Ai Wallet

  • Deliberately vulnerable MCP tool servers for research.
  • Captures full attacker attribution (IP, UA, referer).
  • Archives attacker-deployed "skills" even after deletion.
  • Logs every MCP server attackers connect or deploy.
  • Canary token system alerts on bait secret access.
  • Exports captures to promptfoo, garak, or analysis-tool formats.

Use cases of Cloud Ai Wallet

  • Studying how attackers social-engineer LLM assistants.
  • Observing attacker tool-chaining and skill deployment in MCP.
  • Testing detection and response for agentic AI honeypots.
  • Collecting capture data for red-team exercise analysis.

FAQ from Cloud Ai Wallet

Are the secrets and data in this server real?

No, all secrets and data are fake bait designed to be accessed. They include canary tokens and placeholders. You must supply your own bait secrets.

Is it safe to run this server in production?

No. Run it only in a disposable, isolated environment you control. The server intentionally allows remote code execution (RCE) via the devtools MCP server.

What data does Cloud Ai Wallet capture?

It captures attribution (IP, user agent, referer), full MCP request bodies, skill archival, raw model input/output, provenance, and canary hits. All data is stored as JSONL in CloudWatch/S3.

Who created this server and for what purpose?

It was created by Ali Zain Zahid for defensive AI security research. The project is for defensive purposes only.

Does the server include any real integrations or production endpoints?

No. The server includes decoy surfaces (e.g., /docs, /swagger, /mcp) and MCP bait tool servers for graph, database, storage, fetch, and devtools. All are intended to mimic real services for research.

评论

AI 与智能体 分类下的更多 MCP 服务器