XMI2SDC

Agentic XMI/UML to SDC4 data model converter.

XMI2SDC converts enterprise UML models exported from Sparx Enterprise Architect, MagicDraw, and other tools into SDC4-compliant data models. It uses a multi-agent pipeline to parse, analyze, map, and assemble models via the SDCStudio Assembly API.

Built for air-gapped, self-hosted environments. No cloud dependencies required.

Agents

Type Mapping

Requirements

• Python 3.11+
• An LLM endpoint reachable via litellm, a frontier-model API key or a local model, for the LLM-assisted analysis and edge-case mapping steps.
• For assemble only: a reachable SDCStudio Assembly API (the hosted sdcstudio.axius-sdc.com or a self-hosted instance) plus an API key via SDCSTUDIO_API_KEY (or --api-key).

convert (parse + map to a mapping plan) runs fully standalone; only assemble needs SDCStudio access.

Installation

# Install the package and its dependencies
pip install -e .

# Development (tests, linters)
pip install -e ".[dev]"

A requirements.txt mirroring the runtime dependencies is provided to build an environment directly: pip install -r requirements.txt.

Quick Start

Convert an XMI file

# Configure
cp xmi2sdc.example.yaml xmi2sdc.yaml

# Parse and map (produces JSON mapping plan)
xmi2sdc convert model.xmi

# Specify output path
xmi2sdc convert model.xmi -o mapping_plan.json

MCP Mode

Expose toolsets as MCP servers for LLM clients (Claude Code, Cursor, etc.):

xmi2sdc serve --mcp parser
xmi2sdc serve --mcp mapper

ADK Agent Mode

For standalone agent orchestration, a local LLM is required:

ollama pull qwen2.5:7b-instruct
# Agents use ollama/qwen2.5:7b-instruct by default via LiteLLM

CLI Commands

xmi2sdc convert FILE       # Parse XMI and produce SDC4 mapping plan
xmi2sdc serve --mcp AGENT  # Start MCP stdio server for an agent toolset
xmi2sdc info               # Display config summary and agent inventory
xmi2sdc validate-config    # Validate configuration file
xmi2sdc audit show         # Display audit log with filters

Configuration

Create xmi2sdc.yaml:

sdcstudio:
  base_url: "http://localhost:8000"
  api_key: "${SDC_API_KEY}"

cache:
  root: ".xmi2sdc-cache"

audit:
  path: ".xmi2sdc-cache/audit.jsonl"
  log_level: "standard"  # or "verbose"

output:
  directory: "./output"

Environment variables in ${VAR} syntax are substituted at load time. Missing variables cause an immediate error (fail closed).

Supported Dialects

• Generic XMI 2.x -- Standard OMG XMI exports
• Sparx Enterprise Architect -- Auto-detected via sparxsystems.com namespace
• MagicDraw / Cameo -- Auto-detected via nomagic.com namespace

Dialect is detected automatically from namespace declarations in the XMI file. All dialects normalize to the same ParsedXmiModel intermediate representation.

Docker

docker build -t xmi2sdc .

# MCP server mode
docker run -v $(pwd)/xmi2sdc.yaml:/home/sdc/xmi2sdc.yaml:ro \
    -e XMI2SDC_AGENT=parser \
    xmi2sdc

# CLI mode
docker run -v $(pwd)/xmi2sdc.yaml:/home/sdc/xmi2sdc.yaml:ro \
    xmi2sdc info

Development

Branch Workflow

• main: Production code (protected, PR required)
• dev: Active development

All work happens on dev. Create a Pull Request to merge into main.

Testing

pip install -e ".[dev]"
pytest

# With coverage
pytest --cov=xmi2sdc --cov-report=term-missing

Linting

ruff check src/ tests/
black --check src/ tests/

Architecture

XMI File ──> Parser Agent ──> ParsedXmiModel ──> Mapper Agent ──> Mapping Plan JSON
                                                                        |
                                                                        v
                               Conversion Report <── Reviewer <── Assembler ──> SDCStudio API

The pipeline separates deterministic operations (parsing, type mapping) from LLM-assisted operations (semantic analysis, edge-case resolution). If any stage fails, earlier results are cached and the pipeline can resume without re-processing.

Contributing

See CONTRIBUTING.md for development guidelines and contribution workflow.

Security

See SECURITY.md for the security policy and vulnerability reporting process.

Related Projects

• SDCStudio — the SDC4 data model platform and Assembly API provider that XMI2SDC targets.
• SDC repositories — the open-source SDC ecosystem (validators, agents, and tooling).

License

Apache License 2.0. Copyright 2026 Axius SDC, Inc. See LICENSE and NOTICE.