GlassBox-AutoML-Agent/mcp.json at dev · Gyro0/GlassBox-AutoML-Agent · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
{
  "name": "glassbox-automl",
  "version": "0.1.0",
  "description": "Transparent NumPy-only AutoML, exposed as an IronClaw MCP tool.",
  "runtime": {
    "type": "python",
    "python_version": ">=3.11",
    "command": ["python", "-m", "glassbox.agent.mcp_server"]
  },
  "tools": [
    {
      "name": "auto_fit",
      "description": "Run end-to-end AutoML on a local CSV dataset. Pass the CSV as an on-disk path (csv_path) visible to the local MCP server. Returns a JSON report with EDA summary, model leaderboard, the best model, its hyperparameters, the cross-validation score, and feature importances.",
      "entrypoint": "glassbox.agent.mcp_tool:auto_fit_tool",
      "input_schema": {
        "type": "object",
        "properties": {
          "csv_path": {
            "type": "string",
            "description": "Filesystem path to a local CSV, for example data/_uploaded.csv."
          },
          "target_column": {
            "type": "string",
            "description": "Header name of the target column to predict."
          },
          "task": {
            "type": "string",
            "enum": ["auto", "classification", "regression"],
            "default": "auto"
          },
          "search": {
            "type": "string",
            "enum": ["random", "grid"],
            "default": "random"
          },
          "time_budget": {
            "type": "integer",
            "minimum": 1,
            "default": 120,
            "description": "Soft wall-clock cap (seconds) for random search."
          }
        },
        "required": ["csv_path", "target_column"]
      },
      "output_schema": {
        "type": "object",
        "description": "AutoFit report. The MCP server returns this object directly on success; failures surface as MCP error responses. The CLI shim at glassbox.agent.mcp_tool wraps the same payload as {\"ok\": true, \"report\": ...} or {\"ok\": false, \"error\": ...}.",
        "properties": {
          "task": {"type": "string", "enum": ["classification", "regression"]},
          "best_model": {"type": "string"},
          "best_params": {"type": "object"},
          "cv_score": {"type": "number"},
          "score_summary": {"type": "object"},
          "feature_names": {"type": "array", "items": {"type": "string"}},
          "feature_importances": {"type": "object"},
          "top_features": {"type": "array"},
          "explanation": {"type": "array", "items": {"type": "string"}},
          "eda_summary": {"type": "object"},
          "search_results": {"type": "array"},
          "candidate_models": {"type": "array", "items": {"type": "string"}},
          "processed_shape": {"type": "array", "items": {"type": "integer"}}
        },
        "required": ["task", "best_model", "cv_score", "explanation"]
      }
    }
  ],
  "permissions": {
    "filesystem": ["read"],
    "network": []
  }
}