Jij MCP Server

Instructions

Implementation Reference

• jij_mcp/mcp_setting.py:99-110 (registration)

  Registration of the 'jm_check' MCP tool. Thin handler that invokes the core checker function from jm_checker.
  @mcp.tool() def jm_check(code: str) -> dict: """ Check the code for JijModeling rules. Args: code (str): The code to check. Returns: dict: The result of the check. """ return jijmodeling_check(code)
• jij_mcp/jm_checker.py:57-89 (handler)

  Core handler logic for jm_check: detects 'for' loops (forbidden in JijModeling), executes code in PythonREPL to catch errors, returns diagnostic dict.
  def jijmodeling_check(code_string: str) -> dict: """ Pythonコード文字列をJijModelingのルールに従ってチェックする関数 Args: code_string (str): 解析対象のPythonコード文字列 Returns: dict: チェック結果を含む辞書 """ # for文の検出 for_loop_detected = detect_for_loop(code_string) if for_loop_detected: return { "for_loop_detected": True, "message": _jm_for_statement_check, } # PythonREPLを使用してコードを実行し、エラーをキャッチ result = PythonREPL.run(code_string) if result["status"] == "error": return { "for_loop_detected": False, "message": _jm_for_statement_check, "error": result["error"], } return { "for_loop_detected": False, "message": "No for loop detected and no errors found.", }
• jij_mcp/jm_checker.py:91-119 (helper)

  Helper function to detect 'for' loops using AST parsing (primary) or regex (fallback).
  def detect_for_loop(code_string): """ Pythonコード文字列からfor文の存在を検出する関数 Args: code_string (str): 解析対象のPythonコード文字列 Returns: bool: for文が存在する場合はTrue、存在しない場合はFalse """ # 方法1: ASTを使用した解析（構文的に正しいコードの場合） try: tree = ast.parse(code_string) for node in ast.walk(tree): if isinstance(node, ast.For): return True except SyntaxError: # 構文エラーがある場合は正規表現による検出に進む pass # 方法2: 正規表現を使用した解析（ASTが適用できない場合のバックアップ） # for文の正規表現パターン: 'for'の後にスペースまたはタブ、その後に変数名、'in'が続く形式 pattern = r"\bfor\s+[a-zA-Z_][a-zA-Z0-9_]*\s+in\b" if re.search(pattern, code_string): return True return False
• jij_mcp/jm_checker.py:6-54 (helper)

  Helper string containing guidance and examples on avoiding Python 'for' loops in JijModeling using jm.Element and jm.sum.
  _jm_for_statement_check = """In JijModeling, you cannot use Python loops directly. Instead, you should use the Element objects. # How to write summation in JijModeling without Python loops. The wrong code: ```python objective = 0 for l in range(n_l): for t in range(n_t): for p in range(n_p): for q in range(n_p): if p != q: objective += ChangeCost[p, q] * Switch[p, q, l, t] ``` The correct code: ``` l = jm.Element("l", belongs_to=range(0, n_l)) t = jm.Element("t", belongs_to=range(0, n_t)) p = jm.Element("p", belongs_to=range(0, n_p)) q = jm.Element("q", belongs_to=range(0, n_p)) objective = jm.sum([l, t, p, (q, p != q)], ChangeCost[p, q] * Switch[p, q, l, t]) ``` # How to write forall in JijModeling without Python loops. The wrong code: ```python for l in range(n_l): for t in range(n_t): problem += jm.Constraint( "SingleProductPerLine", jm.sum([X[p, l, t] for p in range(n_p)]) <= 1, forall=[] ) ``` The correct code: ```python l = jm.Element("l", belongs_to=range(0, n_l)) t = jm.Element("t", belongs_to=range(0, n_t)) p = jm.Element("p", belongs_to=range(0, n_p)) problem += jm.Constraint( "SingleProductPerLine", jm.sum(p, X[p, l, t]) <= 1, forall=[l, t] ) ``` """
• jij_mcp/mcp_setting.py:100-110 (schema)

  Input schema: code (str). Output: dict with keys like 'for_loop_detected', 'message', optional 'error'.
  def jm_check(code: str) -> dict: """ Check the code for JijModeling rules. Args: code (str): The code to check. Returns: dict: The result of the check. """ return jijmodeling_check(code)