code-review-graph is a locally-based code intelligence graph engine released by developer Tirth Kanani on June 15, 2026. This tool leverages Tree-sitter (a parsing technology) to parse the source code of various programming languages and precisely extracts only the call and dependency relationships between classes, functions, and variables within the system, constructing it as an in-memory and disk-based Knowledge Graph using SQLite. The constructed Knowledge Graph supports Anthropic's MCP (Model Context Protocol) standard and CLI integration, enabling next-generation AI IDEs and automation agents like Claude Code or Cursor to navigate the codebase without reading the entire file unnecessarily, instead receiving only the relevant relationship information and snippets. In traditional large-scale source code analysis and code review tasks, AI LLMs often wasted tokens and experienced hallucination issues by directly inputting hundreds of files into the context window or relying on simple keyword-based RAG (Retrieval-Augmented Generation). To illustrate, if previous AI coding tools were inefficient, like reading an entire library to find the answer to a question, code-review-graph is structured like first creating a table of contents and a three-dimensional index for the library and then accurately finding only the necessary paragraphs. By transforming the dependency structure of the codebase into structured graph data, it uses a clear node-tracking method instead of random text search when analyzing code changes or the impact of calling a specific function, thereby reducing the LLM context input token count by up to 49 times, even in large monorepo environments. The utility of code-review-graph is also very high in bioinformatics software development and bio data pipeline engineering practices. For example, you can apply git hook-based incremental indexing to a Python and C++ codebase where genomic data analysis algorithms, pipeline modules, and omics data parsing utilities are intricately intertwined across dozens of files. When a developer modifies the source code and creates a PR (Pull Request), calling the code-review-graph` CLI allows the modified function and its associated parent classes and child dependency nodes to be retrieved from the SQLite graph in just a few milliseconds. This enables the AI code review agent to perfectly evaluate the impact of genomic algorithm modifications with only a precise context of less than 1,000 tokens, without scanning the entire library.

code-review-graph is a local-based code intelligence graph engine released by developer Tirth Kanani on June 15, 2026. This tool leverages Tree-sitter (a Tree-sitter parser) technology to parse the source code of various programming languages and precisely extract only the call and dependency relationships between classes, functions, and variables within the system, constructing an in-memory and disk-based Knowledge Graph in SQLite format. The constructed Knowledge Graph is used with Anthropic's MCP (Model Context Pro).

Genomic Analysis Pipeline Monorepo Impact Analysis: In a large-scale pipeline project containing dozens of omics analysis scripts and C++ C-extension modules, when a specific input/output parser is modified, only the top-level analysis modules affected are precisely extracted through graph relationship queries, enabling impact assessment execution without token consumption.