AI Copilot & Model Context Protocol (MCP)

Architectural Overview: Hub + Relay

WebCAE implements a multi-client Model Context Protocol bridge featuring an autonomous Hub + Relay topology:

┌─────────────────────────────────────────────────────────────┐
│                       LLM CLIENTS                           │
│  ┌───────────────────────────┐ ┌──────────────────────────┐ │
│  │   Claude Desktop / Code   │ │     Cursor / Antigravity │ │
│  └─────────────┬─────────────┘ └────────────┬─────────────┘ │
└────────────────┼────────────────────────────┼───────────────┘
          stdio (JSON-RPC)             stdio (JSON-RPC)
                 │                            │
                 ▼                            ▼
        ┌───────────────────┐        ┌───────────────────┐
        │  mcp-bridge (HUB) │◄───────┤ mcp-bridge (RELAY)│
        │  ws://127.0.0.1:  │  ws    │  (no EADDRINUSE   │
        │       8765        │ relay  │   port conflicts) │
        └─────────▲─────────┘        └───────────────────┘
                  │
                  │ WebSocket (bidirectional JSON)
                  ▼
        ┌─────────────────────────────────────────────────┐
        │                WebCAE Application               │
        │  ┌───────────────────────────────────────────┐  │
        │  │ mcpBridgeConnector (auto-reconnect, FIFO) │  │
        │  └─────────────────────┬─────────────────────┘  │
        │                        ▼                        │
        │  ┌───────────────────────────────────────────┐  │
        │  │ window.cae API (CAD, Mesh, Physics, FEM)  │  │
        │  └───────────────────────────────────────────┘  │
        └─────────────────────────────────────────────────┘

Key Architectural Strengths: - Zero Port Collisions: The first MCP server process binds to port 8765 as HUB. Subsequent instances (e.g. multiple Claude windows or Cursor running in parallel) automatically join as RELAY clients, forwarding requests seamlessly. If the HUB shuts down, a RELAY gracefully takes over the port. - Session Slot Management: The bridge guarantees that exactly one active WebCAE browser or desktop window processes simulation commands, putting inactive tabs into a non-interfering standby state. - Strict Entity Addressing Contract: In compliance with WebCAE design standards, all boundary condition targets use verified numerical indices (surface number N, edge number N). Forbidden axis-face identifiers (like +X, bbox_top) are rejected at the schema level to eliminate AI hallucinations.

End-to-End Autonomous Simulation Workflow

Here is the standard conversational workflow between a human engineer, an AI assistant, and the WebCAE MCP bridge:

1. User Request: *"Analyze this steel mounting bracket under a 4.5 kN downward load applied to the top face, with the left end fixed."* 2. Topology Inspection: The AI calls cae_inspect_surfaces() to extract all face IDs, unit normal vectors, and centroids. It identifies the left face (normal [-1, 0, 0], surface #2) and top face (normal [0, 0, 1], surface #1). 3. Material Assignment: The AI calls cae_set_material({ preset: "steel" }) configuring $E = 210\text{ GPa}$, $\nu = 0.3$, $\sigma_y = 250\text{ MPa}$. 4. Boundary Conditions: - cae_add_boundary_condition({ targetNumbers: 2, type: "fix" }) - cae_add_boundary_condition({ targetNumbers: 1, type: "force", value: 4500, component: "z", fz: -4500 }) 5. FEA Solve & Safety Factor: The AI invokes cae_solve(). WebCAE automatically meshes the volume with quadratic tetrahedra (Tet10), solves the equilibrium equations, and returns: - Peak Von Mises stress: $\sigma_{\text{vM}}^{\max} = 142.5\text{ MPa}$ - Max displacement: $u^{\max} = 0.384\text{ mm}$ - Factor of Safety: $\text{FOS} = \frac{250}{142.5} = 1.75$ (SAFE) 6. Visual Inspection: The AI calls cae_screenshot() and embeds the 3D stress contour into the chat for human verification.