dockerbro
MCP server for managing Docker containers, images, and Compose services via natural language, with configurable tool permissions and safety annotations.
README
DockerBro
DockerBro is a proper MCP-compliant Docker management server. Speaks JSON-RPC 2.0 over stdin/stdout and talks to your local Docker daemon.
Works with Zed, Claude, Cursor, VS Code, and any other MCP-compatible AI editor.
Runs on macOS, Linux, and Windows.
pip install dockerbro
Features
| Tool | What it does | Safety |
|---|---|---|
list_containers |
List all containers with status, image, ports | Read-only |
inspect_container |
Detailed container info (network, mounts, env) | Read-only |
start_container |
Start a stopped container | Reversible |
stop_container |
Stop a running container | Reversible |
restart_container |
Restart a container | Reversible |
remove_container |
Remove a container (with optional force) | ⚠️ Destructive |
logs_container |
Get recent log output | Read-only |
exec_container |
Run a command inside a running container | ⚠️ Destructive |
list_images |
List local Docker images | Read-only |
pull_image |
Pull an image from a registry | Reversible |
run_container |
Run a new container with ports, env, etc. | Reversible |
remove_image |
Remove a local image | ⚠️ Destructive |
docker_compose_ps |
List Compose services | Read-only |
docker_compose_up |
Start Compose services | Reversible |
docker_compose_down |
Stop and remove Compose services | ⚠️ Destructive |
docker_compose_logs |
Get Compose service logs | Read-only |
Pre-approving tools
Each tool is annotated with MCP spec annotations (readOnlyHint, destructiveHint, idempotentHint) so spec-aware clients (Claude Code, Cursor, etc.) can auto-approve read-only tools without prompting.
You can also restrict tools at the server level via environment variables, enforced before anything reaches the client:
| Env var | Effect |
|---|---|
DOCKER_MCP_ALLOW_TOOLS |
Comma-separated allowlist — only these tools are exposed and callable |
DOCKER_MCP_DENY_TOOLS |
Comma-separated denylist — always blocked (deny wins over allow) |
Example: only safe read-only tools
Pass the env var in your Zed config (add to the env object):
"env": {
"DOCKER_MCP_ALLOW_TOOLS": "list_containers,inspect_container,logs_container,list_images,docker_compose_ps,docker_compose_logs"
}
If a tool is not pre-approved, the server returns an error:
Tool 'remove_container' is not pre-approved (blocked by server allow/deny config).
Note: Client-side approval (e.g. Zed's
agent.tool_permissions) and server-side pre-approval are independent. The server-side list controls which tools are visible and callable at all; the client controls which of those require a confirmation prompt.
Requirements
- Python 3.9+
- Docker installed and running
Install
Option A: From PyPI (recommended — all platforms)
| Platform | Command |
|---|---|
| macOS / Linux | pip3 install dockerbro |
| Windows | pip install dockerbro |
Any (if pip isn't on PATH) |
python3 -m pip install dockerbro |
command not found: pip? On macOS and many Linux distros, Python 3 installs it aspip3. On Windows it's usuallypiporpy -m pip. When in doubt,python3 -m pip install dockerbroworks everywhere.
This puts a dockerbro command on your PATH, so your editor config needs no file paths at all.
Verify it works:
dockerbro --version 2>/dev/null; echo '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-06-18","capabilities":{},"clientInfo":{"name":"test","version":"1.0"}}}' | dockerbro
Option B: Run from source
git clone https://github.com/ramanailearning-cpu/dockerbro.git
cd dockerbro
pip install -e .
Option C: Run in a Docker container
Build it:
docker build -t dockerbro .
Run it — the socket mount differs per platform:
macOS / Linux:
docker run --rm -i -v /var/run/docker.sock:/var/run/docker.sock dockerbro
Windows (PowerShell):
docker run --rm -i -v //./pipe/docker_engine://./pipe/docker_engine dockerbro
Connecting to the Docker daemon
DockerBro auto-detects your daemon, so this usually needs zero configuration:
| Platform | Default endpoint |
|---|---|
| Linux | /var/run/docker.sock |
| macOS (Docker Desktop) | /var/run/docker.sock, or ~/.docker/run/docker.sock on 4.13+ |
| Windows | Named pipe //./pipe/docker_engine |
To point at a different or remote daemon, set DOCKER_HOST in your config's env block:
"env": { "DOCKER_HOST": "tcp://192.168.1.50:2375" }
macOS Docker Desktop note: if you get a connection error, enable Settings → Advanced → Allow the default Docker socket, or set
DOCKER_HOSTtounix:///Users/YOUR_NAME/.docker/run/docker.sock.
Zed Configuration
Add this to your Zed settings file (Cmd-Shift-P → Open Settings on macOS, Ctrl-Shift-P on Linux/Windows):
"context_servers": {
"dockerbro": {
"command": "dockerbro",
"env": {
"DOCKER_MCP_LOG": "/tmp/dockerbro.log"
}
}
}
Restricting tools server-side
Optionally, add env vars to limit which tools are exposed:
"env": {
"DOCKER_MCP_LOG": "/tmp/dockerbro.log",
"DOCKER_MCP_DENY_TOOLS": "remove_container,remove_image,exec_container,docker_compose_down"
}
Reload Zed (Cmd-Shift-P on macOS / Ctrl-Shift-P on Linux or Windows → Reload Window). The agent will have 16 Docker tools available.
Windows note: If the
dockerbrocommand isn't found by Zed, use the full path shown bywhere dockerbroin your terminal.
Claude Desktop Configuration
macOS
Edit ~/Library/Application Support/Claude/claude_desktop_config.json:
{
"mcpServers": {
"dockerbro": {
"command": "dockerbro",
"env": {
"DOCKER_MCP_LOG": "/tmp/dockerbro.log"
}
}
}
}
Linux
Edit ~/.config/Claude/claude_desktop_config.json:
{
"mcpServers": {
"dockerbro": {
"command": "dockerbro",
"env": {
"DOCKER_MCP_LOG": "/tmp/dockerbro.log"
}
}
}
}
Windows
Edit %APPDATA%\Claude\claude_desktop_config.json (paste the path into Explorer's address bar):
{
"mcpServers": {
"dockerbro": {
"command": "dockerbro",
"env": {
"DOCKER_MCP_LOG": "C:\\Users\\YOU\\dockerbro.log"
}
}
}
}
Testing
Via terminal (pipe JSON-RPC)
# Initialize
echo '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-06-18","capabilities":{},"clientInfo":{"name":"test","version":"1.0"}}}' | dockerbro
# List tools
echo '{"jsonrpc":"2.0","id":2,"method":"tools/list","params":{}}' | dockerbro
# Call a tool
echo '{"jsonrpc":"2.0","id":3,"method":"tools/call","params":{"name":"list_containers","arguments":{"all":true}}}' | dockerbro
Via MCP Inspector
npx @modelcontextprotocol/inspector dockerbro
Architecture
AI Editor (Zed / Claude / Cursor / VS Code)
| (stdin/stdout: JSON-RPC 2.0)
v
dockerbro (Python)
| (Docker SDK)
v
Docker Daemon
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。