selectel-mcp
An MCP server that gives an AI agent controlled access to a Selectel cloud account: OpenStack cloud servers, S3 object storage, and account billing.
README
selectel-mcp
An MCP server that gives an AI agent (Claude Code, Claude Desktop, etc.) controlled access to a Selectel cloud account: OpenStack cloud servers, S3 object storage, and account billing.
What it can do
| Area | Tools |
|---|---|
| Deploy (one-shot) | deploy_docker_app, destroy_app |
| Compute (OpenStack) | list_servers, get_server, list_flavors, list_images, create_server, server_action, delete_server |
| Keypairs | list_keypairs, import_keypair, delete_keypair |
| Security groups | list_security_groups, create_security_group, add_security_group_rule, delete_security_group |
| Network | list_networks, list_subnets, list_floating_ips, create_floating_ip, attach_floating_ip, release_floating_ip |
| Volumes | list_volumes, create_volume, attach_volume, delete_volume |
| Object storage (S3) | list_buckets, list_objects, create_bucket, upload_object, download_object, delete_object |
| Billing | get_balance, get_balance_prediction |
| Account | list_projects |
Paid tools (create_server, deploy_docker_app, create_floating_ip) and destructive
tools (delete_*, destroy_app, release_floating_ip) are gated: the destructive ones
require confirm=True and otherwise return a dry-run preview, so the agent can show you
what would happen first.
Deploy an app in one call
deploy_docker_app(
name="myapp",
image="Ubuntu 22.04",
flavor="SL1.1-1024",
git_repo="https://github.com/you/myapp", # cloned to /opt/app, `docker compose up -d`
ports=[22, 80, 443],
confirm=True, # paid: provisions a server + public IP
)
# → creates a security group, a cloud-init server that installs Docker and runs your
# repo, and a floating IP. destroy_app("myapp") removes it all later.
How auth maps to Selectel
| Layer | Endpoint | Credential |
|---|---|---|
| OpenStack (servers/networks/volumes) | cloud.api.selcloud.ru/identity/v3 |
IAM service user (username + password + account id) → Keystone token |
| Balance / billing | api.selectel.ru |
account IAM token, or a static token (X-Token) for detailed reports |
| Projects | api.selectel.ru/vpc/resell/v2 |
same as above |
| S3 object storage | s3.<pool>.storage.selcloud.ru |
S3 access key issued to the service user |
Setup
- Create a service user in the control panel: Account → Users → Service users. Give it the roles you want the agent to have (start with read-only roles, or scope it to a single project).
- Issue an S3 key to that service user (Service users → Access → S3 keys) if you want object-storage access.
- (Optional) Create a static token (Profile → Access → API keys) for billing reports/transactions.
python3 -m venv .venv
.venv/bin/python -m pip install -e .
cp .env.example .env # then fill in credentials
Fill .env:
SEL_ACCOUNT_ID=123456 # domain name, top-right in the control panel
SEL_USERNAME=mcp-agent # service user
SEL_PASSWORD=...
SEL_PROJECT_ID=<project uuid> # default project (or discover via list_projects)
SEL_REGION=ru-2
SEL_S3_ACCESS_KEY=...
SEL_S3_SECRET_KEY=...
SEL_STATIC_TOKEN= # optional
Run / connect to Claude Code
# from this directory
claude mcp add selectel -- /absolute/path/to/.venv/bin/python -m selectel_mcp
Or run it directly over stdio:
.venv/bin/python -m selectel_mcp
For Claude Desktop, add to claude_desktop_config.json:
{
"mcpServers": {
"selectel": {
"command": "/absolute/path/to/.venv/bin/python",
"args": ["-m", "selectel_mcp"]
}
}
}
The server reads credentials from .env in its working directory (or from real
environment variables).
Safety notes
- The service user is the blast radius — scope its IAM roles to exactly what you want the agent to touch. Read-only roles make the whole server read-only.
.envholds secrets and is git-ignored. Don't commit it.- For infrastructure changes, consider managing them with Terraform (provider
selectel/selectel) so every change goes through a reviewableplan→apply. This MCP server is best for queries and quick actions.
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。