The model is too big to share
A real SysML project won't fit in a prompt. Copy-pasting fragments loses the structure that matters — the relationships, allocations, and traceability.
JG Systems Consulting Ltd. · CATIA Magic MCP
Let your AI work on the SysML model it could never reach.
Your SysML v1 model lives inside CATIA Magic — and your AI assistant can't touch it. You can't paste a 100,000-element model into a chat window, and you certainly can't send a proprietary system design to the cloud. The JGS bridge brings the AI to the model instead: it exposes your live project to Claude Code or any MCP agent over a local, air-gapped connection — ~115 tools to query, audit, and edit the real model in plain English. Nothing leaves your machine.
§01 — The problem
MBSE models grow into tens of thousands of elements. Navigating them, auditing coverage, and keeping requirements traceable is slow manual work — and the one tool that could help, your AI assistant, is locked out. Two walls stand in the way, and most "AI for engineering" stories ignore both.
The model is too sensitive to upload
System designs are IP, often export-controlled. Sending them to a cloud LLM is a non-starter for defense, aerospace, and regulated programmes.
§02 — How it works
A Java plugin runs inside CATIA Magic and exposes the live model as MCP tools over a local HTTP connection. Your agent calls those tools directly — reading and (with a PRO licence) writing the actual model, in its own v1-native vocabulary.
~115 model tools
Find, inspect, navigate, audit, trace requirements, and edit blocks, ports, requirements, and diagrams — the operations engineers actually do.
Plain-English driven
Ask your agent in normal language; it maps the request to the right tools and works on the live model, not a stale export.
v1-native
Speaks SysML v1 vocabulary throughout — Blocks, IBDs, flow ports, allocations — so results match the model you already have.
§03 — Why air-gapped matters
This is the difference that lets you actually use it at work. The bridge binds to
127.0.0.1 only, validates its licence offline, and makes no outbound network
calls. Your model never leaves the workstation — so the security review is short, and the
answer to "where does our IP go?" is "nowhere."
Local only
Loopback connection; no cloud service, no telemetry, no data export.
Offline licence
Ed25519 signature checked locally — no licence server to call home to.
Your tools, your rules
Runs beside the CATIA Magic you already trust. Nothing new to expose to the network.
§04 — Install
1
Install the server
pip install ./server, then create your MCP client config from examples/.mcp.json.example (see docs/install.md).
2
Place the plugin & restart
Copy plugin/ into your CATIA Magic user plugins directory and restart MSOSA. This step runs on your desktop app — an agent cannot do it for you.
3
Verify
Call the ping tool. If it answers, the bridge is live. See configuration.md if not.
# install the server (from the repo root), then place the plugin and restart MSOSA
pip install ./server
Prefer your AI agent do it? The README carries a paste-in install prompt.
§05 — Licence tiers
Start free with read-only access — no licence file. Upgrade when you want the AI to change the model, not just read it.
FREE
Read & audit
Query, navigate, audit, and report on the live model. No licence file required — the fastest way to prove the value.
PRO
Author & edit
Full write access: create and modify blocks, ports, requirements, relationships, and diagrams through your agent. Licence file required.
ENTERPRISE
Script & scale
Everything in PRO plus dangerous writes and Groovy scripting for power workflows. Licence file required.
Upgrade
Ready to let the AI build, not just browse?
The free tier reads your model. A PRO licence lets your agent author and refactor it — create requirements, wire allocations, fix findings, draft diagrams — at the speed you can describe them. ENTERPRISE adds scripting for repeatable, large-scale changes.