Kratos MCP Server

Appearance

Cantilever stiffness and mesh studies ​

Start with the cantilever case. It must contain ProjectParameters.json, Materials.json, and mesh.mdpa, with DISPLACEMENT VTK output enabled. The bundled case uses a 1 m × 0.2 m plate, fixed on the left and loaded on the right. StructuralMechanicsApplication and LinearSolversApplication must be available in the configured build.

When working from this repository, copy the bundled case to your work directory:

bash
cp -r src/kratos_mcp/examples/cantilever /tmp/cantilever-study

The following are MCP tool arguments. Replace /tmp/cantilever-study if you used a different absolute path.

Sweep Young's modulus ​

Call study_start:

json
{
  "case_dir": "/tmp/cantilever-study",
  "max_concurrency": 2,
  "specification": {
    "kind": "parameter",
    "axes": [{
      "file": "Materials.json",
      "pointer": "/properties/0/Material/Variables/YOUNG_MODULUS",
      "values": [105000000000, 210000000000, 420000000000]
    }]
  },
  "responses": [{
    "name": "tip_y",
    "variable": "DISPLACEMENT",
    "point": [1.0, 0.0, 0.0],
    "component": 1,
    "reduction": "final",
    "max_distance": 1e-8
  }]
}

Use the returned study_id with study_status or study_results. All three responses should be negative; doubling stiffness halves the displacement magnitude. Each variant has a distinct child job, snapshot, and execution directory. The original case is unchanged. A solver failure appears alongside successful variants rather than discarding their results.

Refine the mesh ​

Call study_start again with the same case and response, replacing specification with:

json
{
  "kind": "mesh",
  "mesh": {
    "kind": "rectangle",
    "size": [1.0, 0.2]
  },
  "divisions": [[4, 1], [8, 2], [16, 4]]
}

These meshes have 10, 27, and 85 nodes and preserve the named left/right/top/bottom boundaries. The tip response changes with resolution. The difference between 8 × 2 and 16 × 4 should be smaller than between 4 × 1 and 16 × 4; this is checked by the real-build integration test. It does not establish mesh independence by itself. Choose tolerances appropriate to the response and physical problem.

study_results includes previous-level and finest-level differences. To export scalar results, pass csv_file with a new absolute filename, for example /tmp/cantilever-mesh-responses.csv. Full histories and sampled coordinates stay available through the JSON response.

Disconnect, recover, or cancel ​

Closing the MCP connection leaves the coordinator running. Reconnect and use study_list to discover existing studies. If a coordinator crashes, its study becomes interrupted; study_resume verifies the stored inputs and adopts any still-running child jobs before continuing the queue. Use study_cancel to stop both pending work and active jobs while keeping completed results.