docker-mcp
A read-only MCP server that exposes your Docker daemon, allowing AI assistants to list containers, inspect their details, and fetch logs without write access.
README
docker-mcp
A read-only MCP server for Docker. Exposes your Docker daemon over the Model Context Protocol so AI assistants can inspect containers and logs without any write access.
Built with FastAPI and FastMCP.
Tools
| Tool | Description |
|---|---|
list_containers |
List running containers (pass all=true to include stopped) |
get_container |
Full details for a container — ports, networks, env vars, mounts, resource limits |
get_logs |
Fetch container logs with filtering by tail/head, time range, and regex pattern |
get_logs parameters
| Parameter | Default | Description |
|---|---|---|
container_id |
— | Container ID or name |
tail |
100 |
Last N lines. Mutually exclusive with head. |
head |
— | First N lines. Mutually exclusive with tail. |
since |
— | UTC datetime — only return logs after this time |
until |
— | UTC datetime — only return logs before this time |
last_minutes |
— | Convenience shorthand: logs from the last N minutes |
pattern |
— | Regex (or substring) filter applied to log lines |
timestamps |
true |
Prefix each line with an RFC3339 timestamp |
Quick start
Docker run
docker run -d \
--name docker-mcp \
-p 8000:8000 \
-v /var/run/docker.sock:/var/run/docker.sock:ro \
-e MCP_API_KEY=mysecret \
codeberg.org/jhot/docker-mcp:latest
Docker Compose
services:
docker-mcp:
image: codeberg.org/jhot/docker-mcp:latest
ports:
- "8000:8000"
volumes:
- /var/run/docker.sock:/var/run/docker.sock:ro
environment:
MCP_API_KEY: ${MCP_API_KEY:-}
restart: unless-stopped
MCP_API_KEY=mysecret docker compose up -d
Configuration
| Variable | Default | Description |
|---|---|---|
MCP_API_KEY |
unset | Bearer token for /mcp. Unset = no auth. |
DOCKER_HOST |
unset | Docker daemon socket. Unset = /var/run/docker.sock. |
HOST |
0.0.0.0 |
Uvicorn bind address |
PORT |
8000 |
Host port (container always listens on 8000) |
LOG_LEVEL |
info |
Uvicorn log level |
MCP endpoint
The MCP Streamable HTTP endpoint is at http://localhost:8000/mcp.
Configure your client (e.g. Claude Desktop, Cursor) to connect to it:
{
"mcpServers": {
"docker": {
"url": "http://localhost:8000/mcp",
"headers": {
"Authorization": "Bearer mysecret"
}
}
}
}
Omit headers if MCP_API_KEY is not set.
Security
The Docker socket grants full control of the Docker daemon. Treat access to this server the same as access to the socket itself:
- Set
MCP_API_KEYfor any network-accessible deployment. - Use a firewall or reverse proxy to restrict which clients can reach port 8000.
- The
:rosocket mount prevents replacing the socket file but does not restrict Docker API calls —MCP_API_KEYis the primary access control.
Remote Docker
To connect to a remote Docker daemon instead of the local socket, remove the volumes block and set DOCKER_HOST:
docker run -d \
--name docker-mcp \
-p 8000:8000 \
-e DOCKER_HOST=tcp://192.168.1.10:2376 \
-e MCP_API_KEY=mysecret \
codeberg.org/jhot/docker-mcp:latest
Routes
| Path | Auth | Description |
|---|---|---|
/ |
public | Landing page |
/health |
public | {"status": "ok"} |
/api/tools |
public | Tool metadata list |
/mcp |
gated | FastMCP Streamable HTTP endpoint |
Development
Requirements: mise, Python 3.12, Docker
mise exec -- task install # uv sync
mise exec -- task fmt # ruff format
mise exec -- task lint # ruff check
mise exec -- task test # unit tests (no Docker required)
mise exec -- task test-integration # integration tests (requires live Docker)
mise exec -- task run # dev server at http://localhost:8000
mise exec -- task build # docker build
mise exec -- task up # docker compose up --build
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。