Cloud Agent
A hosted AI software engineer that writes code, opens PRs, reviews code, generates tests, runs security scans, and answers codebase questions. Connect from any MCP client (Claude Code, Cursor, Windsurf, or your own agents) and delegate engineering tasks.
README
mcp-server-cloud-agent
MCP server for Cloud Agent — a hosted AI software engineer that writes code, opens PRs, reviews code, generates tests, runs security scans, and answers codebase questions.
Connect from any MCP client (Claude Code, Cursor, Windsurf, or your own agents) and delegate engineering tasks.
How it works
This package is a local stdio MCP proxy that forwards requests to the Cloud Agent hosted backend at cloudagent.metaltorque.dev. Your MCP client communicates with this server over stdio; the server makes authenticated HTTPS calls to the backend on your behalf.
Data flow: MCP client → (stdio) → this server → (HTTPS) → Cloud Agent backend → GitHub
What data leaves your machine:
- Task descriptions, repo names, file paths, and PR URLs you provide
- Your API key (sent over HTTPS only, never over HTTP)
What the backend does with your data:
- Clones repos from GitHub using its own GitHub App credentials
- Executes tasks in isolated sandboxes
- Opens PRs and posts reviews to GitHub on your behalf
Tools (9)
| Tool | Description | Side effects |
|---|---|---|
run_task |
Write code, fix bugs, add features — returns result + PR URL | Creates branches and PRs |
review_pr |
Review a GitHub PR with structured feedback | Optionally posts comments to GitHub |
ask_codebase |
Ask questions about any GitHub repo (auto-indexes on first use) | Read-only |
generate_tests |
Generate tests for a file, opens a PR | Creates branches and PRs |
security_scan |
Security + dependency scan across one or more repos | Read-only |
list_sessions |
List recent sessions with status, cost, duration, PR URLs | Read-only |
list_playbooks |
List available workflow templates | Read-only |
run_playbook |
Run a playbook against a repo | Creates branches and PRs |
get_usage |
Usage stats — sessions, cost, time saved, breakdowns | Read-only |
Setup
1. Get an API key
Sign in to your Cloud Agent workspace at cloudagent.metaltorque.dev and generate an API key at /auth/api-key. Keys use the ca_* prefix.
2. Configure your MCP client
Claude Code (~/.claude.json):
{
"mcpServers": {
"cloud-agent": {
"command": "npx",
"args": ["-y", "mcp-server-cloud-agent"],
"env": {
"CLOUD_AGENT_API_KEY": "ca_your_key_here"
}
}
}
}
Cursor / Windsurf (MCP settings):
{
"mcpServers": {
"cloud-agent": {
"command": "npx",
"args": ["-y", "mcp-server-cloud-agent"],
"env": {
"CLOUD_AGENT_API_KEY": "ca_your_key_here"
}
}
}
}
Environment Variables
| Variable | Required | Description |
|---|---|---|
CLOUD_AGENT_API_KEY |
Yes | API key (ca_* prefix) from your Cloud Agent workspace |
CLOUD_AGENT_URL |
No | Backend URL (defaults to https://cloudagent.metaltorque.dev) |
Usage Examples
Once configured, your MCP client can call these tools directly:
Fix a bug:
"Use cloud-agent run_task on myorg/myapp to fix the broken login flow"
Review a PR:
"Use cloud-agent review_pr on https://github.com/myorg/myapp/pull/42"
Ask about code:
"Use cloud-agent ask_codebase on myorg/myapp: how does authentication work?"
Generate tests:
"Use cloud-agent generate_tests on myorg/myapp for src/auth.ts"
Security scan:
"Use cloud-agent security_scan on myorg/myapp and myorg/api"
Sample Output
run_task response:
{
"response": "Fixed the login redirect bug. Changed src/auth.ts to properly handle OAuth callback URLs.",
"cost_usd": 0.42,
"duration_ms": 45000,
"pr_url": "https://github.com/myorg/myapp/pull/87"
}
security_scan response:
{
"repos_scanned": 1,
"vulnerabilities": 3,
"secrets_found": 0,
"findings": [...]
}
Troubleshooting
| Problem | Solution |
|---|---|
| "CLOUD_AGENT_API_KEY is required" | Set the env var in your MCP client config |
| "Refusing to send API key over insecure HTTP" | Use HTTPS (the default). Don't set CLOUD_AGENT_URL to an HTTP URL |
| "Request timed out" | Tasks can take up to 10 minutes. Check list_sessions for status |
| "HTTP 401" | Your API key is invalid or expired. Generate a new one |
| "HTTP 429" | Rate limited. Wait and retry |
License
MIT
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。