Knowledge Graph MCP Server

A high-performance, lightweight Knowledge Graph MCP server for code analysis, optimized for Apple Silicon.

Features

• Fast Indexing: Optimized for Apple Silicon (MLX) and multi-core CPUs

• Smart Incremental Indexing: Content hashing skips unchanged files (instant re-indexing)

• Smart Scope: Automatically ignores dependencies (node_modules, venv, target, etc.)

• Hybrid Search: Semantic + exact keyword search

• Graph Analysis: Understand code relationships (definitions, references, calls)

• Multi-Language: Python, JavaScript, TypeScript, Java, Go

Installation

MCP Client Configuration

Add to your MCP client (e.g., Claude Desktop) configuration:

Available Tools

The server exposes 8 tools that LLMs can use for code analysis:

1. reindex_codebase

Purpose: Index a codebase to enable all other tools.

Parameters:

• project_root_path (string): Absolute or relative path to the project directory

LLM Usage:

Returns: Indexing statistics (files processed, nodes/edges created, time taken)

2. search_code

Purpose: Search for code using hybrid semantic + keyword search with optional filtering.

Parameters:

• query (string): Search query (concept or exact name)

• limit (int, default=5): Maximum results to return

• file_type (string, optional): Filter by file extension (e.g., "py", "java", "ts")

LLM Usage:

Returns: List of matching code snippets with:

• File path and line numbers

• Code content

• Match type (semantic or keyword)

• Confidence score (0.0-1.0, where 1.0 = 100% match)

3. get_structure

Purpose: Get the structure of a file (imports, classes, functions, methods).

Parameters:

• file_path (string): Path to the file

LLM Usage:

Returns: Structured JSON with:

• imports: List of import statements

• definitions: Classes and functions with:

  • Nested methods (for classes)

  • Line numbers

  • Docstrings

4. find_definitions

Purpose: Find where a symbol (function, class, variable) is defined.

Parameters:

• symbol_name (string): Name of the symbol to find

LLM Usage:

Returns: List of definitions with file paths, line numbers, and code snippets

5. find_references

Purpose: Find where a symbol is called/used in the codebase.

Parameters:

• name (string): Name of the symbol

LLM Usage:

Returns: List of call sites with:

• Source file and line number

• Caller name

• Target file and type

6. get_file_summary

Purpose: Get a summary of a file's contents.

Parameters:

• file_path (string): Path to the file

LLM Usage:

Returns: File summary with:

• Total lines

• Top-level structure (classes, functions)

7. get_usage_context

Purpose: Get the full context for a symbol (definition + usage).

Parameters:

• symbol_name (string): Name of the symbol

LLM Usage:

Returns: Combined information:

• Where it's defined (with code)

• Where it's used (with surrounding context)

8. validate_edit

Purpose: Validate code syntax before making changes (supports Python, JavaScript, TypeScript, Java, Go).

Parameters:

• file_path (string): Path to the file

• new_content (string): Proposed new content

LLM Usage:

Returns: Validation result:

• valid: true/false

• errors: List of syntax errors (if any)

Example Workflows

Initial Setup

Code Exploration

Refactoring

Performance

• Indexing Speed: ~500 files/s (initial), ~3500 files/s (incremental)

• Search Speed: <100ms for most queries

• Languages: Python, JavaScript, TypeScript, Java, Go

• Smart Scope: Auto-excludes ~3000+ dependency files

License

MIT License - see LICENSE for details