zig-mcp
MCP server for Zig that connects AI coding assistants to ZLS (Zig Language Server) via LSP. Provides 16 tools for code intelligence (hover, go-to-definition, references, completions, diagnostics, rename, format) and build/test operations.
README
zig-mcp
MCP server for Zig that connects AI coding assistants to ZLS via the Language Server Protocol.
Works with Claude Code, Cursor, Windsurf, and any MCP-compatible client.
AI assistant <--(MCP stdio)--> zig-mcp <--(LSP pipes)--> ZLS
|
zig build / test / check
Requirements
Install
Claude Code plugin (recommended)
Install directly from the Claude Code interface — no manual build needed:
# 1. Add the marketplace
/plugin marketplace add nzrsky/zig-mcp
# 2. Install the plugin
/plugin install zig-mcp@zig
Or as a one-liner from the terminal:
claude plugin marketplace add nzrsky/zig-mcp && claude plugin install zig-mcp@zig
The binary is built automatically on first use. Just make sure zig and zls are in your PATH.
Manual build
git clone https://github.com/nzrsky/zig-mcp.git
cd zig-mcp
zig build -Doptimize=ReleaseFast
Binary is at zig-out/bin/zig-mcp.
Setup (manual install only)
If you installed via the plugin system, skip this section — everything is configured automatically.
Claude Code
# add globally
claude mcp add zig-mcp -- /absolute/path/to/zig-mcp --workspace /path/to/your/zig/project
# add for current project only
claude mcp add --scope project zig-mcp -- /absolute/path/to/zig-mcp --workspace /path/to/your/zig/project
Or edit ~/.claude/mcp_servers.json:
{
"mcpServers": {
"zig-mcp": {
"command": "/absolute/path/to/zig-mcp",
"args": ["--workspace", "/path/to/your/zig/project"]
}
}
}
If you omit
--workspace, zig-mcp uses the current working directory.
Cursor
Add to .cursor/mcp.json in your project:
{
"mcpServers": {
"zig-mcp": {
"command": "/absolute/path/to/zig-mcp",
"args": ["--workspace", "/path/to/your/zig/project"]
}
}
}
Windsurf
Add to ~/.codeium/windsurf/mcp_config.json:
{
"mcpServers": {
"zig-mcp": {
"command": "/absolute/path/to/zig-mcp",
"args": ["--workspace", "/path/to/your/zig/project"]
}
}
}
Options
--workspace, -w <path> Project root directory (default: cwd)
--zls-path <path> Path to ZLS binary (default: auto-detect from PATH)
--help, -h Show help
--version Show version
Tools
Code intelligence (via ZLS)
| Tool | What it does |
|---|---|
zig_hover |
Type info and docs for a symbol |
zig_definition |
Go to definition |
zig_references |
Find all references |
zig_completion |
Completion suggestions |
zig_diagnostics |
Errors and warnings for a file |
zig_format |
Format a file |
zig_rename |
Rename a symbol across the workspace |
zig_document_symbols |
List all symbols in a file |
zig_workspace_symbols |
Search symbols across the project |
zig_code_action |
Quick fixes and refactors for a range |
zig_signature_help |
Function signature at cursor |
Build & run
| Tool | What it does |
|---|---|
zig_build |
Run zig build with optional args |
zig_test |
Run tests (whole project or single file, with optional filter) |
zig_check |
Run zig ast-check on a file |
zig_version |
Show Zig and ZLS versions |
zig_manage |
Manage Zig versions via zvm |
How it works
zig-mcp spawns ZLS as a child process and talks to it over stdin/stdout using the LSP protocol (Content-Length framing). On the other side, it speaks MCP (newline-delimited JSON-RPC) to the AI assistant.
Three threads:
- main -- reads MCP requests, dispatches tool calls, writes responses
- reader -- reads LSP responses from ZLS, correlates by request ID
- stderr -- forwards ZLS stderr to the server log
If ZLS crashes, zig-mcp automatically restarts it and re-opens all tracked documents.
Files are opened in ZLS lazily on first access -- no need to manage document state manually.
Development
# build
zig build
# run tests (~75 unit tests)
zig build test
# run manually
echo '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"capabilities":{}}}' | \
zig-out/bin/zig-mcp --workspace . 2>/dev/null
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。