Codegraphcontext

CodeGraphContext — Structural Code Intelligence

CodeGraphContext (CGC) parses source code with tree-sitter and builds a queryable graph of functions, classes, and their relationships. Use it to answer structural questions that text search can't: "who calls this?", "what would break?", "where is dead code?"

Supports 14 languages: Python, JS/TS, Java, C/C++, C#, Go, Rust, Ruby, PHP, Swift, Kotlin, Dart, Perl.

When to invoke

Use CGC instead of grep when questions are structural:

• Tracing callers / callees across a large codebase
• Impact analysis before refactoring a function or class
• Finding dead code before a cleanup
• Identifying the most complex functions to prioritize
• Answering "what calls X?" or "what does Y depend on?"

Do not use for text/regex search, reading file contents, or single-file edits.

Installation

pip install codegraphcontext
# or with uv:
uv pip install codegraphcontext

Verify: cgc --help

Default backend: KùzuDB (zero-config, embedded). No server needed.

Core workflow

1 — Index the repo

# Index current directory
cgc index .

# Index a specific path
cgc index /path/to/project

Run once; re-run after large changes or use cgc watch for live updates.

2 — Analyze

# Who calls a function?
cgc analyze callers <function_name>

# What does a function call?
cgc analyze calls <function_name>

# Visualize call chain interactively
cgc analyze calls <function_name> --viz

# Find complex functions (default threshold: 10)
cgc analyze complexity --threshold 15

# Detect dead code
cgc analyze dead-code

# Explore class hierarchy
cgc analyze tree <ClassName> --viz

3 — Find patterns

cgc find pattern "<search_term>"

4 — List indexed repos

cgc list

5 — Live watch (keep graph current)

cgc watch /path/to/project

MCP server mode (optional)

CGC can expose its graph as an MCP server so Claude can query it conversationally:

cgc mcp setup    # interactive wizard
cgc mcp start    # launch the server

When the MCP server is running, Claude can use cgc.* tools directly without shell commands. Prefer the CLI above unless you need persistent conversational access.

Typical task patterns

Impact analysis before refactoring:

cgc index .
cgc analyze callers target_function

Find what to clean up:

cgc analyze dead-code
cgc analyze complexity --threshold 20

Understand an unfamiliar module:

cgc analyze calls entry_point_function --viz
cgc analyze tree CoreClassName --viz

Integration with other skills

• repomix — use repomix for broad file-content context; use CGC for structural/relational queries
• mcp-knowledge-graph — CGC focuses on code structure; knowledge-graph stores architectural decisions and patterns

Notes

• First cgc index may take 30–60 seconds on large repos
• Graph is stored locally in .cgc/ by default
• Add .cgc/ to .gitignore