go-unifi-mcp

go-unifi-mcp

MCP server for UniFi Network Controller enabling AI assistants to manage UniFi infrastructure via natural language. It supports firewall rules, IPv6, and uses lazy/eager tool modes to minimize context usage.

Category
访问服务器

README

go-unifi-mcp

A Model Context Protocol (MCP) server for UniFi Network Controller, written in Go.

Overview

go-unifi-mcp provides an MCP interface to UniFi Network Controller, enabling AI assistants and other MCP clients to interact with your UniFi infrastructure.

Why this exists

I couldn’t find an MCP server that supported both v1 and v2 firewall rules and IPv6, so I built one. This wraps the go-unifi library (which I trust from my Terraform provider experience) and leans on its generated API surface. The server is generated from the controller’s own API definitions, which makes it much easier to keep tool coverage up to date as UniFi evolves.

UniFi controller versioning

This project generates tools against the same UniFi Controller version pinned by go-unifi. When go-unifi updates its supported controller version, we regenerate our field definitions and tool metadata to match. We support the same controller range; see their controller support range.

Installation

Binary (GitHub Releases)

Download pre-built binaries from the Releases page. Binaries are available for macOS and Linux (amd64/arm64).

# macOS (Apple Silicon)
curl -L https://github.com/claytono/go-unifi-mcp/releases/latest/download/go-unifi-mcp_darwin_arm64.tar.gz | tar xz
sudo mv go-unifi-mcp /usr/local/bin/

# macOS (Intel)
curl -L https://github.com/claytono/go-unifi-mcp/releases/latest/download/go-unifi-mcp_darwin_amd64.tar.gz | tar xz
sudo mv go-unifi-mcp /usr/local/bin/

# Linux (amd64)
curl -L https://github.com/claytono/go-unifi-mcp/releases/latest/download/go-unifi-mcp_linux_amd64.tar.gz | tar xz
sudo mv go-unifi-mcp /usr/local/bin/

# Linux (arm64)
curl -L https://github.com/claytono/go-unifi-mcp/releases/latest/download/go-unifi-mcp_linux_arm64.tar.gz | tar xz
sudo mv go-unifi-mcp /usr/local/bin/

Homebrew

Available from the claytono/homebrew-tap tap:

brew install claytono/tap/go-unifi-mcp

Nix

# Run without installing
nix run github:claytono/go-unifi-mcp

# Install to your profile
nix profile install github:claytono/go-unifi-mcp

Docker

Multi-architecture images (amd64/arm64) are published to GitHub Container Registry.

# Latest (pinned to most recent release, rebuilt on base image updates)
docker pull ghcr.io/claytono/go-unifi-mcp:latest

# Edge (built from main on every merge, unstable)
docker pull ghcr.io/claytono/go-unifi-mcp:edge

Go Install

go install github.com/claytono/go-unifi-mcp/cmd/go-unifi-mcp@latest

Configuration

UniFi Credentials

The server requires access to a UniFi Network Controller. Two authentication methods are supported:

  1. API Key (preferred): Create an API key in your UniFi controller under Settings > Control Plane > Integrations. Set UNIFI_HOST and UNIFI_API_KEY.

  2. Username/Password: Use a local admin account. Set UNIFI_HOST, UNIFI_USERNAME, and UNIFI_PASSWORD.

Claude Desktop

Add to your claude_desktop_config.json:

Using the binary:

{
  "mcpServers": {
    "unifi": {
      "command": "/usr/local/bin/go-unifi-mcp",
      "env": {
        "UNIFI_HOST": "https://your-controller:443",
        "UNIFI_API_KEY": "your-api-key"
      }
    }
  }
}

Using Docker:

{
  "mcpServers": {
    "unifi": {
      "command": "docker",
      "args": [
        "run",
        "-i",
        "--rm",
        "-e",
        "UNIFI_HOST",
        "-e",
        "UNIFI_API_KEY",
        "ghcr.io/claytono/go-unifi-mcp:latest"
      ],
      "env": {
        "UNIFI_HOST": "https://your-controller:443",
        "UNIFI_API_KEY": "your-api-key"
      }
    }
  }
}

Claude Code

claude mcp add unifi -- go-unifi-mcp

Then set the required environment variables in your shell before running claude.

Environment Variables

Variable Required Default Description
UNIFI_HOST Yes — UniFi controller URL
UNIFI_API_KEY * — API key (preferred auth method)
UNIFI_USERNAME * — Username for password auth
UNIFI_PASSWORD * — Password for password auth
UNIFI_SITE No default UniFi site name
UNIFI_VERIFY_SSL No true Whether to verify SSL certs
UNIFI_LOG_LEVEL No error go-unifi client log level
UNIFI_TOOL_MODE No lazy Tool registration mode

* Either UNIFI_API_KEY or both UNIFI_USERNAME and UNIFI_PASSWORD must be set.

Log Levels

The UNIFI_LOG_LEVEL variable controls logging from the underlying go-unifi client library. The default is error because the client otherwise emits INFO messages to stderr, which can interfere with tools like mcp-cli that parse JSON on stdout/stderr.

Level Description
disabled No logging
trace Most verbose, including wire details
debug Debug messages
info Informational messages
warn Warnings only
error Errors only (default)

Tool Modes

The server supports two tool registration modes, following the pattern established by unifi-network-mcp:

Mode Tools Context Size Description
lazy 3 ~200 tokens Meta-tools only (default, recommended for LLMs)
eager 252 ~55K tokens All tools registered directly

Lazy mode (default) registers only 3 meta-tools that provide access to 252 UniFi operations (generated from the controller API):

  • tool_index - Search/filter the tool catalog by category or resource
  • execute - Execute any tool by name with arguments
  • batch - Execute multiple tools in parallel

This dramatically reduces context window usage while preserving full functionality. The LLM first queries the index to find relevant tools, then executes them via the dispatcher.

Eager mode registers all 252 tools directly, which may be useful for non-LLM clients or debugging but consumes significant context.

Update semantics: Updates use a read-modify-write flow against the controller API. We fetch the current resource, merge your fields, and submit the full object. This avoids clearing unspecified fields, but it is not atomic and concurrent updates can race (last write wins) because the UniFi API does not expose etags or revision IDs. In practice this is unlikely to be an issue, but it's something to be aware of.

ID Resolution

Responses from the UniFi API contain opaque ID references (e.g. network_id, usergroup_id, networkconf_id). By default, the server resolves these to human-readable names by looking up the referenced resource and injecting a sibling _name field:

{
  "src_networkconf_id": "609fbf24e3ae433962e000de",
  "src_networkconf_name": "IOT"
}

Resolution uses a per-request cache, so listing 100 firewall rules that reference networks only makes one additional ListNetwork API call. Typical overhead is 10-40ms depending on how many distinct resource types are referenced.

To disable resolution for a specific call, pass "resolve": false in the tool arguments.

Query Parameters

All list operations support optional post-processing parameters for filtering and projecting results.

filter — Match items by field values. Supports three operators:

// Exact match
{"filter": {"type": "usw"}}

// Substring match (case-insensitive)
{"filter": {"name": {"contains": "office"}}}

// Regular expression (RE2 syntax: https://github.com/google/re2/wiki/Syntax)
{"filter": {"name": {"regex": "^ap-.*"}}}

// Multiple conditions (ANDed together)
{"filter": {"type": "uap", "name": {"contains": "echo"}}}

search — Case-insensitive full-text search across all string field values:

{ "search": "living room" }

fields — Project the response to include only specific keys:

{ "fields": ["name", "ip", "mac"] }

Parameters can be combined. Execution order is filter → search → fields, so you can filter on fields that are excluded from the output:

{
  "filter": { "type": "uap" },
  "search": "echo",
  "fields": ["name", "ip"]
}

Development

Prerequisites

  • Nix with flakes enabled
  • direnv (optional but recommended)

Developing

# Clone the repository
git clone https://github.com/claytono/go-unifi-mcp.git
cd go-unifi-mcp

# Enter the development environment
nix develop
# Or with direnv:
direnv allow

# Install pre-commit hooks
pre-commit install

# Run linters
task lint

# Run tests
task test

# Run tests with coverage
task coverage

Available Tasks

task lint        # Run linters via pre-commit
task test        # Run tests
task coverage    # Run tests with coverage checks
task build       # Build the binary
task generate    # Run go generate

Testing with mcp-cli

The development environment includes mcp-cli for interactive testing of the MCP server.

  1. Create .envrc.local with your UniFi credentials (not tracked in git):

    export UNIFI_HOST="https://your-controller:443"
    export UNIFI_API_KEY="your-api-key"
    # Or use username/password:
    # export UNIFI_USERNAME="admin"
    # export UNIFI_PASSWORD="password"
    
  2. Build the binary:

    task build
    
  3. Test with mcp-cli:

    The .mcp_servers.json config provides two server entries:

    • go-unifi-mcp - eager mode (252 tools)
    • go-unifi-mcp-lazy - lazy mode (3 meta-tools)

    Eager mode (direct tool access):

    # List tools (shows all 252)
    mcp-cli info go-unifi-mcp
    
    # Call a tool directly
    mcp-cli call go-unifi-mcp list_device '{}'
    mcp-cli call go-unifi-mcp list_network '{"site": "default"}'
    

    Lazy mode (meta-tools):

    # List tools (shows only 3 meta-tools)
    mcp-cli info go-unifi-mcp-lazy
    
    # Query the tool index
    mcp-cli call go-unifi-mcp-lazy tool_index '{}'
    mcp-cli call go-unifi-mcp-lazy tool_index '{"category": "list"}'
    mcp-cli call go-unifi-mcp-lazy tool_index '{"resource": "network"}'
    
    # Execute a tool via the dispatcher
    mcp-cli call go-unifi-mcp-lazy execute '{"tool": "list_device", "arguments": {}}'
    
    # Batch execute multiple tools
    mcp-cli call go-unifi-mcp-lazy batch '{"calls": [{"tool": "list_network", "arguments": {}}, {"tool": "list_device", "arguments": {}}]}'
    

Credits

This project builds upon:

  • go-unifi - Go client library for UniFi Network Controller
  • unifi-network-mcp - Python MCP server for UniFi that inspired this project
  • mcp-go - Go SDK for Model Context Protocol

License

MPL-2.0

推荐服务器

Baidu Map

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。

官方
精选
JavaScript
Playwright MCP Server

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。

官方
精选
TypeScript
Magic Component Platform (MCP)

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。

官方
精选
本地
TypeScript
Audiense Insights MCP Server

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。

官方
精选
本地
TypeScript
VeyraX

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。

官方
精选
本地
graphlit-mcp-server

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。

官方
精选
TypeScript
Kagi MCP Server

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。

官方
精选
Python
e2b-mcp-server

e2b-mcp-server

使用 MCP 通过 e2b 运行代码。

官方
精选
Neon MCP Server

Neon MCP Server

用于与 Neon 管理 API 和数据库交互的 MCP 服务器

官方
精选
Exa MCP Server

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。

官方
精选