CodeGG

Local-first code intelligence graph for AI coding agents

Zero-config · Zero external APIs · 100% local

What is CodeGG?

CodeGG is a local-first code intelligence engine that parses your codebase with tree-sitter, builds a SQLite knowledge graph of every symbol, call, and relationship, and exposes it to AI coding agents over MCP (Model Context Protocol).

Instead of your agent spending time grepping, globbing, and reading files to understand your code, CodeGG answers structural questions instantly:

• "How does a request reach the database?"
• "What calls this function?"
• "What would change X break?"
• "Where is this symbol defined?"

One codegg_explore call returns the relevant source, the call flow, and the impact radius — usually with zero file reads.

How it works

Your Agent (Claude Code / Cursor / Codex CLI / opencode)
    │  "How does login work?"
    │  calls codegg_explore → returns source + flow + impact
    ▼
CodeGG MCP Server
    │
    ▼
SQLite knowledge graph (100% local)
    symbols · edges · files · FTS5 full-text search

Pipeline

1. Extraction — tree-sitter parses source into ASTs. Per-language queries extract nodes (functions, classes, methods) and edges (calls, imports, inheritance).
2. Storage — everything goes into a local SQLite database (.codegg/codegg.db) with FTS5 full-text search.
3. Resolution — references are resolved: function calls → definitions, imports → files, framework routing patterns, and dynamic dispatch bridges.
4. Auto-sync — the MCP server watches your project with native OS file events. Changes are debounced and incrementally synced. No configuration needed.

Quick Start

1. Install

npm i -g @jonusnattapong/codegg

2. Wire up your agents

codegg install

Auto-detects Claude Code, Cursor, Codex CLI, and opencode — configures their MCP servers automatically.

3. Index a project

cd your-project
codegg init -i

That's it. Open your agent and ask a structural question.

Key Features

| | | |---|---| | One-tool answers | codegg_explore returns source + call flow + impact in a single call | | Full-text search | FTS5-powered symbol search across the entire codebase | | Impact analysis | Trace callers, callees, and transitive dependencies before changing code | | Auto-sync | Native file watcher — the graph stays fresh as you code | | 20+ languages | TypeScript, Python, Go, Rust, Java, C#, PHP, Ruby, C/C++, Swift, Kotlin, Dart, and more | | Framework-aware | Django, FastAPI, Spring, Express, NestJS, Gin, Rails, Laravel, ASP.NET, React Router, and more | | Dynamic dispatch | Bridges React render, JSX children, Vue/Svelte templates, Django signals, FastAPI Depends, event emitters, callback observers, and more | | 100% local | No data leaves your machine. No API keys. No external services |

CLI Reference

codegg                         # Interactive installer
codegg install                 # Run installer
codegg uninstall               # Remove CodeGG from agents
codegg init [path]             # Initialize project (--index to also index)
codegg uninit [path]           # Remove CodeGG from a project
codegg index [path]            # Full index (--force to re-index)
codegg sync [path]             # Incremental update
codegg status [path]           # Show index stats
codegg query <search>          # Search symbols
codegg files [path]            # Show file structure
codegg callers <symbol>        # Find callers of a symbol
codegg callees <symbol>        # Find what a symbol calls
codegg impact <symbol>         # Analyze impact of changes
codegg serve --mcp             # Start MCP server

MCP Tools

When running as an MCP server, CodeGG exposes:

| Tool | What it does | |---|---| | codegg_explore | Primary tool. Returns relevant source + call flow + impact in one call | | codegg_search | Search by symbol name | | codegg_callers | Find what calls a symbol | | codegg_callees | Find what a symbol calls | | codegg_impact | Transitive impact analysis | | codegg_node | Full source for one symbol (returns every overload) | | codegg_files | Indexed file structure | | codegg_status | Index health |

Supported Languages

TypeScript, JavaScript, Python, Go, Rust, Java, C#, PHP, Ruby, C, C++, Objective-C, Swift, Kotlin, Scala, Dart, Lua, Luau, Svelte, Vue, Liquid, Pascal/Delphi.

Supported Agents

• Claude Code
• Cursor
• Codex CLI
• opencode

Why CodeGG?

AI agents are powerful, but exploring an unfamiliar codebase costs them tokens on every grep/Read/glob. CodeGG pre-indexes everything so your agent answers structural questions with one MCP call instead of dozens of file reads — cutting tool calls by ~58% and cost by ~16% on typical architecture questions.

License

MIT