CallGraph MCP Server

MCP server for analyzing call graphs in Axelor projects.

What is this?

A Model Context Protocol (MCP) server that lets Claude Code intelligently query your project's call graph:

• Find where methods/classes are used
• Analyze change impact
• Navigate dependencies
• Trace call chains across Java and XML files

Quick Start

1. Install dependencies

pip install -r requirements.txt

2. Configure Claude Desktop

Edit your claude_desktop_config.json:

Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "callgraph": {
      "command": "python",
      "args": ["C:/Users/nicolasv/MCP_servers/CallGraph/mcp_callgraph_server.py"]
    }
  }
}

3. Extract your project's call graph

cd /path/to/your/axelor/project
python /path/to/CallGraph/run_extraction.py . --mode full

This creates .vector-raw-db/ in your project directory.

4. Restart Claude Desktop

The server auto-detects the database in your current working directory.

Architecture

run_extraction.py           # CLI entry point
    ↓
ExtractionManager          # Orchestrates extraction
    ↓
├── JavaASTExtractor       # Extracts Java call graph (via JavaASTService)
├── TypeScriptASTExtractor # Extracts TypeScript call graph
├── AxelorXmlExtractor     # Extracts XML references
├── AxelorRepoManager      # Downloads Axelor dependencies
└── StorageWriter          # Writes to ChromaDB

Query Path:
mcp_callgraph_server.py    # MCP server
    ↓
StorageReader              # Reads from ChromaDB
    ↓
ChromaDB (.vector-raw-db)  # Vector database

Extraction Modes

Full Mode (Recommended)

python run_extraction.py /path/to/project --mode full

• Auto-detects Axelor versions from gradle.properties / settings.gradle
• Downloads axelor-open-platform and axelor-open-suite
• Caches Axelor extractions (reused across projects)
• Extracts local modules

Local Mode (Fast)

python run_extraction.py /path/to/project --mode local

• Only extracts /modules directory
• Faster for iterative development
• Doesn't update Axelor caches

Options

--reset true/false   # Reset database before extraction (default: true for full, false for local)
--limit N            # Limit entries per repo (for testing)

MCP Tools

find_usages

Find all usages of a symbol with optional recursion.

find_usages(symbol="validateMove", depth=2)

impact_analysis

Analyze the impact of changing a method.

impact_analysis(symbol="computeTotal", depth=3, only_custom=True)

get_definition

Find where a symbol is defined.

get_definition(symbol="MoveValidateService")

find_callers / find_callees

Navigate the call graph.

find_callers(symbol="validateMove")
find_callees(symbol="processMove")

search_by_file

Find all usages in a specific file.

search_by_file(file_path="MoveValidateService.java")

get_stats

Database statistics.

get_stats()  # Global stats
get_stats(module="open-auction-base")  # Module stats

extract

Trigger extraction from Claude Code.

extract(mode="full", reset=True)

File Naming Conventions

The codebase follows PascalCase file naming where each file name matches its main class:

JavaASTExtractor.py         → class JavaASTExtractor
TypeScriptASTExtractor.py   → class TypeScriptASTExtractor
AxelorXmlExtractor.py       → class AxelorXmlExtractor
AxelorRepoManager.py        → class AxelorRepoManager
ExtractionManager.py        → class ExtractionManager
StorageWriter.py            → class StorageWriter (Write operations)
StorageReader.py            → class StorageReader (Read operations)

This follows modern conventions (similar to Java/C#/TypeScript) for better IDE navigation.

How It Works

Auto-detection

The MCP server automatically detects the database in your current working directory:

cwd = Path.cwd()  # Claude Code's current project

if (cwd / ".vector-semantic-db").exists():
    db_path = cwd / ".vector-semantic-db"  # Prioritize semantic
elif (cwd / ".vector-raw-db").exists():
    db_path = cwd / ".vector-raw-db"

Example:

• Open C:\project-a\ in Claude Code → Uses C:\project-a\.vector-raw-db\
• Open C:\project-b\ in Claude Code → Uses C:\project-b\.vector-raw-db\

Extraction Process

1. Detect Axelor versions from settings.gradle / gradle.properties
2. Download dependencies (axelor-open-platform, axelor-open-suite)
3. Extract to caches (reused across projects with same version)
4. Copy caches to project database
5. Extract local modules (your custom code)

What Gets Extracted

Java (via JavaASTService):

• Method calls
• Constructor calls
• Field access
• Inheritance (extends/implements)
• Annotations
• Method/class declarations

XML (via AxelorXmlExtractor):

• Action methods
• Domain class/field references
• View references
• Menu actions

TypeScript (via TypeScriptASTExtractor):

• Function calls
• React component usage
• Imports/exports
• Hook calls

Cache System

Axelor dependencies are cached per version:

axelor-repos/
├── axelor-open-platform-7.2.3/
│   └── .vector-raw-db/          # Cached extraction
└── axelor-open-suite-7.2.3/
    └── .vector-raw-db/

Benefits:

• Extract Axelor once per version
• Share across all projects using same version
• Faster subsequent extractions

Performance

Operation	Time	Size
Full extraction (3000 files)	~5-10 min	~150 MB
Local extraction (500 files)	~1-2 min	~50 MB
find_usages (depth=0)	10-50ms	-
impact_analysis (depth=3)	200ms-1s	-

Troubleshooting

No database found

# Extract first
cd /path/to/project
python /path/to/CallGraph/run_extraction.py . --mode full

JavaASTService not starting

The service starts automatically but requires Java:

# Check Java installation
java -version

# Manual start (for debugging)
cd Extracteurs/JavaASTService
./gradlew service

Empty results

# Check database stats
get_stats()

# Re-extract with reset
extract(mode="full", reset=True)

Development

Project Structure

CallGraph/
├── mcp_callgraph_server.py    # MCP server entry point
├── run_extraction.py           # CLI entry point
├── Extracteurs/
│   ├── ExtractionManager.py   # Orchestrator (has main())
│   ├── JavaASTExtractor.py    # Java extraction (pure lib)
│   ├── TypeScriptASTExtractor.py
│   ├── AxelorXmlExtractor.py
│   ├── AxelorRepoManager.py
│   ├── StorageWriter.py       # ChromaDB writes
│   ├── StorageReader.py       # ChromaDB reads
│   └── JavaASTService/        # Java parser service
└── axelor-repos/              # Cached Axelor repos

Running Tests

# Test extraction
python run_extraction.py /path/to/project --mode local --limit 100

# Test specific extractor (now pure libraries - use ExtractionManager)
cd Extracteurs
python ExtractionManager.py --project-root /path/to/project --local

Design Patterns

CQRS (Command Query Responsibility Segregation):

• StorageWriter: All write operations (insert, batch, reset)
• StorageReader: All read operations (query, filter, format)

Pure Libraries:

• Extractors have no main() - they're imported by ExtractionManager
• Single orchestrator: ExtractionManager.py

Limitations

• One project at a time (based on cwd)
• No hot-reload (restart Claude Desktop to switch projects)
• Memory: ~200-500 MB for ChromaDB

License

MIT

Resources

• Extracteurs README
• JavaASTService README
• MCP Protocol
• ChromaDB