pygraph-mcp 0.6.0

AST-based Python/Flask/FastAPI codebase indexer for AI agents

pygraph

AST-based Python/Flask codebase indexer for AI coding agents.

pygraph parses your Python project into a queryable graph (graph.json) using static analysis (AST). It detects Flask routes, blueprints, templates, CLI commands, decorators, dependencies, and more — with zero network calls and no telemetry.

Designed as a companion tool for AI agents (opencode, Cline, etc.) via MCP, or for direct CLI use during development.

---

Quick Start

pip install pygraph-mcp

# Or with uv:
uv pip install pygraph-mcp

Index your project and start querying:

cd my-python-project
pygraph build
pygraph public
pygraph node "my_function"
pygraph callers "my_function"

---

Commands

Command	Description
build	Index project into .pygraph/graph.json
node	Show details of a symbol
callers	Show who calls the given symbol
callees	Show what the given symbol calls
source	Show source code of a file
query	Search symbols by pattern (regex or substring)
context	Show symbol with callers, callees, tests, and source
imports	Show imports of a file
public	List all exported (public) symbols
focus	Show JSON detail for a symbol
impact	Show downstream impact (BFS from symbol)
path	Show shortest call path between two symbols
orphans	List uncalled private symbols
trace	Find error messages and trace their call paths
complexity	Show McCabe cyclomatic complexity
coupling	Show afferent/efferent coupling metrics
hotspot	Show high-risk symbols (complexity x coupling)
deps	List external dependencies
boundaries	Check architecture boundary violations
changes	Show symbol changes since a git ref
stale	List files not modified in N days
plan	Generate a change plan report
review	Generate a Markdown code review report
graph-report	Generate a Markdown report about the codebase
add-opencode-plugin	Create .opencode.json with pygraph MCP config
mcp	Start MCP stdio server for AI agent integration

---

MCP Integration

Start the MCP server and connect your AI agent:

pygraph mcp --root /path/to/project

All query commands are exposed as MCP tools, making them callable from opencode, Cline, and any MCP-compatible client.

---

Flask Support

pygraph detects Flask-specific constructs out of the box:

• Route decorators (@app.route) with HTTP methods and URL parameters
• Blueprint creation and registration (register_blueprint)
• Template rendering (render_template, render_template_string)
• Error handlers (@app.errorhandler)
• CLI commands (@app.cli.command)
• Flask extension usage

---

Architecture Enforcement

Create .pygraph/boundaries.json:

{
  "layers": [
    {"name": "views",     "pattern": "src/views/",     "allowed": ["services"]},
    {"name": "services",  "pattern": "src/services/",  "allowed": ["models", "repos"]},
    {"name": "models",    "pattern": "src/models/",    "allowed": []},
    {"name": "repos",     "pattern": "src/repos/",     "allowed": ["models"]}
  ]
}

Then run:

pygraph boundaries

---

Requirements

• Python 3.11+
• uv (recommended) or pip

---

Development

git clone https://github.com/shvmgyl15/pygraph.git
cd pygraph
uv sync
uv run pytest
uv run mypy src
uv run ruff check

---

How It Works

┌──────────┐    ┌──────────────┐    ┌───────────┐
│  Scanner  │ → │  Extractors  │ → │   Graph   │
│ (walker)  │   │ (AST, Flask) │   │  (JSON)   │
└──────────┘    └──────────────┘    └───────────┘
                                          ↓
                                   ┌──────────┐
                                   │  Query   │
                                   │  Engine  │
                                   └──────────┘

1. Scan — walks the project tree, respects .gitignore, classifies files
2. Extract — parses Python source with ast, extracts symbols, calls, imports, Flask decorators, dependencies
3. Graph — serializes everything to .pygraph/graph.json (incremental builds supported)
4. Query — CLI or MCP tools query the graph for callers, callees, impact, paths, orphans, complexity, coupling, etc.

---

Inspiration

pygraph was inspired by and follows the same design philosophy as:

• tsgraph — TypeScript codebase indexer
• gograph — Go codebase indexer

Both projects use AST-based static analysis to create queryable code graphs for AI coding agents. pygraph adapts the same concept to Python and Flask.