# MCP Integration
How to use Tree-Sitter MCP with Claude and other AI tools via the Model Context Protocol.
## Setup
### Claude Desktop
Add to your Claude Desktop configuration file (`~/.config/claude-desktop/claude_desktop_config.json`):
```json
{
"mcpServers": {
"tree-sitter-mcp": {
"command": "npx",
"args": ["@nendo/tree-sitter-mcp", "--mcp"],
"cwd": "/path/to/your/project"
}
}
}
```
### Other MCP Clients
Tree-Sitter MCP follows the standard MCP protocol and should work with any compliant client:
```bash
tree-sitter-mcp --mcp
```
## Available Tools
When running as an MCP server, Tree-Sitter MCP provides these tools:
### `search_code`
Find code elements by name with fuzzy matching and progressive content inclusion. Automatically includes code content based on result count: single result gets full content, 2-3 results get limited content, 4+ results get metadata only.
### `find_usage`
Trace where functions, classes, and variables are used.
### `analyze_code`
Comprehensive code quality and structure analysis.
### `check_errors`
Find actionable syntax errors with detailed context and fix suggestions.
## Usage Patterns
### Code Exploration
```
You: "Find all React components in this project"
Claude: Uses search_code with types=["function", "class"] and query="Component"
```
### Refactoring Impact Analysis
```
You: "I want to rename the User class. Show me all places it's used."
Claude:
1. Uses search_code to find User class definition
2. Uses find_usage to trace all references
3. Provides comprehensive impact analysis
```
### Code Quality Review
```
You: "Analyze the code quality of this project"
Claude: Uses analyze_code with all analysis types to generate detailed report
```
### Architecture Understanding
```
You: "Help me understand how authentication works in this codebase"
Claude:
1. Uses search_code to find auth-related functions
2. Uses find_usage to trace authentication flow
3. Maps out the authentication architecture
```
## Configuration Options
### Working Directory
Set the `cwd` in your MCP configuration to the project root:
```json
{
"mcpServers": {
"tree-sitter-mcp": {
"command": "npx",
"args": ["@nendo/tree-sitter-mcp", "--mcp"],
"cwd": "/Users/yourname/projects/your-project"
}
}
}
```
### Debug Logging
Enable debug output by setting environment variables:
```json
{
"mcpServers": {
"tree-sitter-mcp": {
"command": "npx",
"args": ["@nendo/tree-sitter-mcp", "--mcp"],
"cwd": "/path/to/project",
"env": {
"TREE_SITTER_MCP_DEBUG": "true"
}
}
}
}
```
### Multiple Projects
Configure different instances for different projects:
```json
{
"mcpServers": {
"frontend-analysis": {
"command": "npx",
"args": ["@nendo/tree-sitter-mcp", "--mcp"],
"cwd": "/path/to/frontend"
},
"backend-analysis": {
"command": "npx",
"args": ["@nendo/tree-sitter-mcp", "--mcp"],
"cwd": "/path/to/backend"
}
}
}
```
## Troubleshooting
### Server Not Starting
1. Verify the command path is correct
2. Check that the working directory exists
3. Ensure you have the package installed globally: `npm list -g @nendo/tree-sitter-mcp`
### Permission Errors
Make sure Claude Desktop has access to your project directory and npm binaries.
### Performance Issues
For very large projects (>100k files), the initial parsing may take a few seconds. Consider:
- Using `.gitignore` patterns to exclude unnecessary files
- Focusing analysis on specific subdirectories
### Debug Information
Enable debug logging to see what the server is doing:
```json
{
"env": {
"TREE_SITTER_MCP_DEBUG": "true"
}
}
```
Check the logs in `~/.config/claude-desktop/logs/` for detailed error information.
## Best Practices
### Effective Prompts
- Be specific about what you're looking for: "Find all API endpoint handlers" vs "Find functions"
- Ask for context: "Show me how the User class is used in authentication"
- Request explanations: "Explain the architecture of the payment system"
### Project Organization
- Keep your project structure clean for better analysis results
- Use meaningful function and class names for better search results
- Follow consistent naming conventions
### Performance Tips
- The tool works best on well-structured codebases
- Large monorepos may take longer to analyze
- Focus searches on specific directories when possible
## Integration Examples
### Code Review Assistant
```
You: "Review the code quality of the new payment module"
Claude:
1. Uses analyze_code on src/payment/ directory
2. Provides quality metrics and suggestions
3. Identifies potential issues and improvements
```
### Documentation Generator
```
You: "Generate documentation for all public APIs"
Claude:
1. Uses search_code to find all exported functions
2. Uses find_usage to understand how they're used
3. Generates comprehensive API documentation
```
### Dependency Mapping
```
You: "Show me all the dependencies of the UserService class"
Claude:
1. Uses search_code to find UserService
2. Uses find_usage to see what it depends on
3. Creates a dependency graph
```
