cognitive-modules 0.4.0__py3-none-any.whl → 0.5.0__py3-none-any.whl

cognitive/mcp_server.py

@@ -0,0 +1,245 @@
+"""
+Cognitive Modules MCP Server
+
+提供 MCP (Model Context Protocol) 接口，让 Claude Code、Cursor 等工具可以使用 Cognitive Modules。
+
+启动方式:
+    cogn mcp
+    
+或直接运行:
+    python -m cognitive.mcp_server
+"""
+
+from typing import Optional, Any
+import json
+import os
+
+try:
+    from mcp.server.fastmcp import FastMCP
+except ImportError:
+    raise ImportError(
+        "MCP dependencies not installed. "
+        "Install with: pip install cognitive-modules[mcp]"
+    )
+
+from .registry import list_modules, find_module
+from .loader import load_module
+from .runner import run_module as execute_module
+
+# ============================================================
+# MCP Server Setup
+# ============================================================
+
+mcp = FastMCP(
+    "Cognitive Modules",
+    description="可验证的结构化 AI 任务规范 - 提供代码审查、任务排序等模块",
+)
+
+
+# ============================================================
+# Tools
+# ============================================================
+
+@mcp.tool()
+def cognitive_run(
+    module: str,
+    args: str,
+    provider: Optional[str] = None,
+    model: Optional[str] = None,
+) -> dict[str, Any]:
+    """
+    运行 Cognitive Module，获取结构化的 AI 分析结果。
+    
+    Args:
+        module: 模块名称，如 "code-reviewer", "task-prioritizer"
+        args: 输入参数，如代码片段或任务列表
+        provider: LLM 提供商（可选），如 "openai", "anthropic"
+        model: 模型名称（可选），如 "gpt-4o", "claude-3-5-sonnet-20241022"
+    
+    Returns:
+        结构化结果，包含分析内容、confidence（置信度）和 rationale（推理过程）
+    
+    Example:
+        cognitive_run("code-reviewer", "def login(u,p): return db.query(f'SELECT * FROM users WHERE name={u}')")
+        
+        返回:
+        {
+            "ok": true,
+            "data": {
+                "issues": [{"type": "security", "severity": "critical", "description": "SQL 注入"}],
+                "confidence": 0.95,
+                "rationale": "检测到字符串拼接 SQL"
+            }
+        }
+    """
+    # 检查模块是否存在
+    module_path = find_module(module)
+    if not module_path:
+        return {"ok": False, "error": f"Module '{module}' not found"}
+    
+    # 设置 provider（如果指定）
+    original_provider = os.environ.get("LLM_PROVIDER")
+    original_model = os.environ.get("LLM_MODEL")
+    
+    try:
+        if provider:
+            os.environ["LLM_PROVIDER"] = provider
+        if model:
+            os.environ["LLM_MODEL"] = model
+        
+        # 执行模块
+        result = execute_module(module, args=args)
+        
+        return {"ok": True, "data": result, "module": module}
+    except Exception as e:
+        return {"ok": False, "error": str(e), "module": module}
+    finally:
+        # 恢复原始环境变量
+        if original_provider:
+            os.environ["LLM_PROVIDER"] = original_provider
+        elif "LLM_PROVIDER" in os.environ and provider:
+            del os.environ["LLM_PROVIDER"]
+        
+        if original_model:
+            os.environ["LLM_MODEL"] = original_model
+        elif "LLM_MODEL" in os.environ and model:
+            del os.environ["LLM_MODEL"]
+
+
+@mcp.tool()
+def cognitive_list() -> dict[str, Any]:
+    """
+    列出所有已安装的 Cognitive Modules。
+    
+    Returns:
+        模块列表，包含名称、位置和格式信息
+    
+    Example:
+        cognitive_list()
+        
+        返回:
+        {
+            "modules": [
+                {"name": "code-reviewer", "location": "builtin", "format": "v1"},
+                {"name": "task-prioritizer", "location": "builtin", "format": "v1"}
+            ],
+            "count": 2
+        }
+    """
+    modules = list_modules()
+    return {
+        "modules": [
+            {"name": m["name"], "location": m["location"], "format": m["format"]}
+            for m in modules
+        ],
+        "count": len(modules),
+    }
+
+
+@mcp.tool()
+def cognitive_info(module: str) -> dict[str, Any]:
+    """
+    获取 Cognitive Module 的详细信息。
+    
+    Args:
+        module: 模块名称
+    
+    Returns:
+        模块详情，包含名称、描述、输入输出格式等
+    
+    Example:
+        cognitive_info("code-reviewer")
+        
+        返回:
+        {
+            "name": "code-reviewer",
+            "version": "1.0.0",
+            "description": "代码安全和质量审查",
+            "responsibility": "分析代码并识别潜在问题"
+        }
+    """
+    module_path = find_module(module)
+    if not module_path:
+        return {"ok": False, "error": f"Module '{module}' not found"}
+    
+    try:
+        m = load_module(module)
+        meta = m.get("meta", {})
+        
+        return {
+            "ok": True,
+            "name": meta.get("name", module),
+            "version": meta.get("version"),
+            "description": meta.get("description") or meta.get("responsibility"),
+            "responsibility": meta.get("responsibility"),
+            "input_schema": m.get("input_schema"),
+            "output_schema": m.get("output_schema"),
+        }
+    except Exception as e:
+        return {"ok": False, "error": str(e)}
+
+
+# ============================================================
+# Resources
+# ============================================================
+
+@mcp.resource("cognitive://modules")
+def get_modules_resource() -> str:
+    """获取所有模块列表（资源形式）"""
+    modules = list_modules()
+    return json.dumps([m["name"] for m in modules], indent=2)
+
+
+@mcp.resource("cognitive://module/{name}")
+def get_module_resource(name: str) -> str:
+    """获取单个模块的 prompt 内容"""
+    module_path = find_module(name)
+    if not module_path:
+        return f"Module '{name}' not found"
+    
+    try:
+        m = load_module(name)
+        return m.get("prompt", "")
+    except Exception as e:
+        return f"Error: {e}"
+
+
+# ============================================================
+# Prompts
+# ============================================================
+
+@mcp.prompt()
+def code_review_prompt(code: str) -> str:
+    """生成代码审查提示"""
+    return f"""请使用 cognitive_run 工具审查以下代码：
+
+```
+{code}
+```
+
+调用: cognitive_run("code-reviewer", "{code[:100]}...")
+"""
+
+
+@mcp.prompt()
+def task_prioritize_prompt(tasks: str) -> str:
+    """生成任务排序提示"""
+    return f"""请使用 cognitive_run 工具对以下任务进行优先级排序：
+
+{tasks}
+
+调用: cognitive_run("task-prioritizer", "{tasks}")
+"""
+
+
+# ============================================================
+# 启动入口
+# ============================================================
+
+def serve():
+    """启动 MCP 服务器"""
+    mcp.run(transport="stdio")
+
+
+if __name__ == "__main__":
+    serve()