GitHub - unslop-xyz/noodles: Your codebase was probably AI-generated. Get a better handle on it. Noodles creates interactive diagrams that visualize how your code actually works, so you can understand what the AI built without reading every line.

3 min read Original article ↗

TL;DR

Your codebase was probably AI-generated. Get a better handle on it.

Noodles creates interactive diagrams that visualize how your code actually works, so you can understand what the AI built without reading every line.

noodles demo

Hosted version available at https://unslop.xyz.

What it does

  • Builds function call graphs using tree-sitter AST parsing
  • Generates mermaid diagrams showing code flow
  • Interactive viewer (pan, zoom, drill down into sub-diagrams)
  • Repo analyzer - Analyze an entire repository
  • PR analyzer - Analyze a GitHub PR to see what changed

Installation

pip install git+https://github.com/unslop-xyz/noodles.git

Or for development:

git clone https://github.com/unslop-xyz/noodles.git
cd noodles
pip install -e .

Configuration

Set your LLM API key as an environment variable:

export ANTHROPIC_API_KEY=your-key-here

Supported providers: anthropic (default), openai, gemini, groq, huggingface. To use a different provider:

export LLM_PROVIDER=openai
export OPENAI_API_KEY=your-key-here

Alternatively, create a .env file in your working directory (see .env.example if you cloned the repo).

An API key enables AI-powered enrichment of node descriptions and edge labels. Without it, the call graph will still be generated but nodes and edges won't have human-readable descriptions.

Noodles uses gh to fetch your repo/PR. To set it up, run:

brew install gh
gh auth login

Usage

Analyze a repository

Analyze a PR

noodles pr <github-pr-url>

Open viewer for existing results

noodles viewer <result-dir>

Options

  • --output-dir <dir> - Custom output directory for results
  • --no-view - Don't open the viewer after analysis

MCP Server (Claude Code Integration)

Noodles can run as an MCP server, allowing Claude Code to analyze PRs and repositories directly.

Setup

  1. Install with MCP dependencies:

    pip install "noodles[mcp]"

  2. Add to your Claude Code settings (~/.claude/settings.json):

    {
  "mcpServers": {
    "noodles": {
      "command": "noodles-mcp"
    }
  }
}

  3. Restart Claude Code

  4. Verify with /mcp - you should see the noodles server listed

Available Tools

Once configured, Claude Code can use these tools:

  • analyze_pr - Analyze a GitHub PR
  • analyze_repo - Analyze an entire repository
  • analyze_local_repo - Analyze a local repository (no cloning)
  • analyze_changes - Analyze uncommitted local changes
  • get_diagram - Get Mermaid diagram from analysis
  • get_call_graph - Get full call graph JSON
  • find_path - Find call path between two functions
  • get_callers - Get functions that call a given function
  • get_callees - Get functions called by a given function
  • filter_graph - Filter nodes by type (entry_points, endpoints, new, updated, orphans)
  • get_summary - Get plain English summary of analysis impact
  • get_changes - Get detailed change info for modified functions

Viewer controls

  • Pan - Click and drag
  • Zoom - Scroll wheel
  • Drill down - Click [+] nodes
  • Back - Escape

Supported languages

The call graph builder uses tree-sitter for AST parsing. Currently supported:

Extension Language
.py Python
.js JavaScript
.jsx JavaScript
.ts TypeScript
.tsx TSX

React/Next.js JSX component usage (<MyComponent />) is detected as function calls, so component hierarchies appear correctly in the call graph.

Not yet supported

  • Other languages - Requires adding tree-sitter grammar and function detection logic

Limitations

Call detection

The call graph is built by detecting function calls in the AST. This works well for:

  • Direct function calls: foo(), obj.method()
  • Imported function calls
  • JSX component usage: <MyComponent />, <UI.Card />

This does not detect:

  • Dynamic calls - getattr(obj, 'method')(), obj[key]()
  • Callbacks passed to frameworks - e.g., route handlers registered via decorators

PR analyzer

The PR analyzer prunes the call graph to functions affected by a PR. This works best when:

  • Changed functions call or are called by other functions
  • The codebase has interconnected function calls

It produces limited results when:

  • Changes are to isolated/standalone functions
  • The codebase uses patterns not detected by AST analysis (decorators, dynamic dispatch)