treesitter-mcp
An MCP server and CLI that uses Tree-sitter to parse and analyze code.
README
Tree-sitter MCP Server & CLI
An MCP server and CLI that uses Tree-sitter to parse and analyze code.
What it does
- Parse files and get their AST
- Extract functions and variables
- Build call graphs
- Find where functions/variables are used
- Run custom Tree-sitter queries
- List imports/includes
- Extract source code for specific line/column ranges
- Use as a standalone CLI or as an MCP server for Claude Desktop
Install
Requires Python 3.10+.
# Install directly
uv pip install treesitter-mcp
# Or install a specific version
uv pip install treesitter-mcp==2.1
# Or clone and install
git clone https://github.com/pwno-io/treesitter-mcp.git
cd treesitter-mcp
uv pip install -e .
This installs both treesitter-mcp and ts-cli entry points.
Or run without installing:
uvx treesitter-mcp
Running
As an MCP server (default)
For use with Claude Desktop or other MCP clients:
treesitter-mcp
See docs/MCP_USAGE.md for how to configure.
Standalone CLI (ts-cli)
Run analysis directly from the terminal:
ts-cli path/to/file.py
ts-cli path/to/file.py --ast --max-depth 2
ts-cli path/to/file.py --find-function main --include-source
ts-cli --supported-languages
Use --output-file to write results to JSON files instead of stdout.
HTTP mode
For testing or manual use:
treesitter-mcp --http --port 8000 --host 127.0.0.1
Limiting tools
Only expose certain tools with --tools:
treesitter-mcp --http --port 8000 --tools treesitter_analyze_file,treesitter_get_ast
Or via URL query param: http://127.0.0.1:8000?tools=treesitter_analyze_file,treesitter_get_ast
Tools available:
treesitter_analyze_file- Basic analysistreesitter_get_ast- Full ASTtreesitter_get_call_graph- Function callstreesitter_find_function- Find function definitionstreesitter_find_variable- Find variablestreesitter_get_source_for_range- Extract source code for a rangetreesitter_get_supported_languages- What's supportedtreesitter_get_node_at_point- AST node at a line/columntreesitter_get_node_for_range- AST node for a rangetreesitter_cursor_walk- Walk tree with contexttreesitter_run_query- Custom Tree-sitter queriestreesitter_find_usage- Find symbol usagestreesitter_get_dependencies- Extract imports/includes
If you don't specify --tools, everything is exposed.
Writing output to file
All tools support an optional output_file parameter. When provided, the tool
writes its result directly to the specified file (as pretty-printed JSON) instead
of returning it. This is useful for large outputs like ASTs that could cause
context overload in agents.
Example:
# Returns result to file, minimal response to agent
treesitter_get_ast(file_path="large_file.py", output_file="~/output/ast.json")
# Returns: {"status": "written", "output_file": "/home/user/output/ast.json", "bytes_written": 123456}
The tool will:
- Automatically create parent directories if they don't exist
- Expand
~to your home directory - Warn (to stderr) if overwriting an existing file
- Return a minimal confirmation dict on success, or an error dict if writing fails
Including source code in results
The treesitter_find_function and treesitter_find_variable tools support an
optional include_source parameter. When set to True, each matched symbol
includes its source code in the result:
treesitter_find_function(file_path="server.py", name="main", include_source=True)
# Returns: {"query": "main", "matches": [{"name": "main", ..., "source": "def main():\n ..."}]}
Language support
| Language | analyze_file | get_ast | get_call_graph | find_function | find_variable | find_usage | get_dependencies |
|---|---|---|---|---|---|---|---|
| C | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
| C++ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
| Python | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
| JavaScript | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
| TypeScript | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
| Go | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
| Java | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
| PHP | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
| Rust | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
| Ruby | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ |
File extensions
| Language | Extensions |
|---|---|
| C | .c |
| C++ | .cpp, .cc, .cxx, .h, .hpp |
| Python | .py |
| JavaScript | .js, .jsx, .mjs, .cjs |
| TypeScript | .ts, .tsx, .cts, .mts |
| Go | .go |
| Java | .java |
| PHP | .php, .phtml |
| Rust | .rs |
| Ruby | .rb |
Docs
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。