clangd-mcp-server
Provides C++ code intelligence tools for AI agents via the Model Context Protocol, enabling symbol navigation, type information, and diagnostics.
README
Clangd MCP Server
Model Context Protocol server for clangd on large C++ codebases.
This MCP provides coding agents like Claude Code with a collection of tools that they may use to answer natural language queries from the user:
find_definition: Jump to symbol definitions- "Find the definition at src/foo.cpp:42:10"
find_references: Find all references to a symbol- "Find all references to the function at bar.h:100"
get_hover: Get type information and documentation- "What's the type at baz.cpp:200:15?"
workspace_symbol_search: Search symbols across workspace- "Find symbols matching 'HttpRequest'"
find_implementations: Find interface/virtual method implementations- "Find implementations of interface.h:50"
get_document_symbols: Get hierarchical symbol tree for a file- "Show all symbols in main.cpp"
get_diagnostics: Get compiler errors, warnings, and notes- "Show errors in src/foo.cpp"
get_call_hierarchy: Get function callers and callees- "Show callers/callees at main.cpp:100:5"
get_type_hierarchy: Get base classes and derived classes- "Show base/derived classes at foo.h:42"
Requirements
- Node.js >= 18.0.0
- clangd
- A C++ project with
compile_commands.json
Installation
# From npm (eventually!)
# npm install -g clangd-mcp-server
# From source
git clone https://github.com/felipeerias/clangd-mcp-server.git
cd clangd-mcp-server
npm install && npm run build && npm link
Configuration
Generating compile_commands.json
CMake: cmake -DCMAKE_EXPORT_COMPILE_COMMANDS=ON /path/to/source
GN (Chromium): gn gen out/Default
gn gen --export-compile-commands out/default
ln -sf out/Default/compile_commands.json .
claude mcp add clangd-mcp-server clangd-mcp-server
Other: Check your project's documentation.
Claude Code Configuration
claude mcp add clangd-mcp-server clangd-mcp-server
Or add manually to ~/.claude.json or .claude.json:
{
"mcpServers": {
"clangd": {
"command": "clangd-mcp-server",
"env": {"PROJECT_ROOT": "/path/to/your/project"},
"alwaysAllow": ["*"]
}
}
}
The alwaysAllow: ["*"] field allows all tools to run without prompting for user approval.
Project-Specific Configuration (CLAUDE.md)
To help Claude Code automatically use clangd MCP tools for your C++ project, add to your project's CLAUDE.md:
## C++ Code Navigation
This project uses the clangd MCP server for C++ code intelligence. Use these tools for:
- Finding definitions and references
- Getting type information
- Searching symbols
- Finding implementations
- Getting diagnostics
Environment Variables
| Variable | Description | Default |
|---|---|---|
PROJECT_ROOT |
Project workspace root | Current directory |
COMPILE_COMMANDS_DIR |
Path to compile_commands.json directory | Auto-detected |
CLANGD_PATH |
Path to clangd binary | Auto-detected |
CLANGD_ARGS |
Additional clangd arguments | Auto-configured |
LOG_LEVEL |
MCP log level (ERROR/WARN/INFO/DEBUG) | INFO |
CLANGD_LOG_LEVEL |
Clangd log level | error |
Clangd auto-detection order: CLANGD_PATH → project bundled (Chromium: third_party/llvm-build/.../clangd) → system PATH
Some large projects bundle their own clangd.
Chromium is auto-detected at third_party/llvm-build/Release+Asserts/bin/clangd.
For other projects in a similar situation, set CLANGD_PATH to specify the bundled clangd.
For bettern performance, background indexing is disabled by default. Usually there is already an axisting clangd server taking care of indexing the codebase. You can enable it with:
{"env": {"CLANGD_ARGS": "--background-index --limit-results=1000"}}
Large projects might consider using remote index.
Verbose logging may be enabled with:
{"env": {"LOG_LEVEL": "DEBUG", "CLANGD_LOG_LEVEL": "verbose"}}
Examples:
// Chromium (auto-detects bundled clangd)
{"mcpServers": {"clangd": {"command": "clangd-mcp-server",
"env": {"PROJECT_ROOT": "/home/user/chromium/src"},
"alwaysAllow": ["*"]}}}
// Custom clangd binary
{"mcpServers": {"clangd": {"command": "clangd-mcp-server",
"env": {"CLANGD_PATH": "/custom/path/clangd"},
"alwaysAllow": ["*"]}}}
// Custom args (e.g., enable background indexing)
{"mcpServers": {"clangd": {"command": "clangd-mcp-server",
"env": {"CLANGD_ARGS": "--background-index --limit-results=1000"},
"alwaysAllow": ["*"]}}}
Architecture
Claude Code
↓ MCP (stdio)
clangd-mcp-server
├── ClangdManager (lifecycle, health monitoring)
├── LSPClient (JSON-RPC over stdio)
├── FileTracker (didOpen/didClose)
└── Tools (find_definition, find_references, etc.)
↓ LSP requests
clangd subprocess
Development
npm install # Install
npm run build # Build
npm run watch # Watch mode
npm test # Run tests
node dist/index.js # Test locally
License
MPL-2.0 - See LICENSE
References
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。