[Developer] #424 add MCP reasoning tab

src/main/java/com/cloudhandson/vpdbackoffice/service/McpReasoningService.java

@@ -0,0 +1,162 @@
+package com.cloudhandson.vpdbackoffice.service;
+
+import com.cloudhandson.vpdbackoffice.domain.mcp.McpReasoningCommand;
+import com.cloudhandson.vpdbackoffice.domain.mcp.McpReasoningResult;
+import com.cloudhandson.vpdbackoffice.domain.mcp.McpToolView;
+import com.cloudhandson.vpdbackoffice.domain.probe.ProbeCommand;
+import com.cloudhandson.vpdbackoffice.domain.probe.ProbeResult;
+import com.cloudhandson.vpdbackoffice.domain.probe.ProbeStatus;
+import com.cloudhandson.vpdbackoffice.domain.protectedobject.ProtectedObject;
+import com.fasterxml.jackson.databind.ObjectMapper;
+import com.fasterxml.jackson.databind.node.ObjectNode;
+import java.util.List;
+import org.springframework.stereotype.Service;
+
+@Service
+public class McpReasoningService {
+
+  private static final int MAX_EVIDENCE_ROWS = 20;
+
+  private final ProtectedObjectService protectedObjectService;
+  private final OrdsProbeService ordsProbeService;
+  private final McpToolRegistry toolRegistry;
+  private final OpenAiCompatibleClient aiClient;
+  private final ObjectMapper objectMapper;
+
+  public McpReasoningService(
+      ProtectedObjectService protectedObjectService,
+      OrdsProbeService ordsProbeService,
+      McpToolRegistry toolRegistry,
+      OpenAiCompatibleClient aiClient,
+      ObjectMapper objectMapper
+  ) {
+    this.protectedObjectService = protectedObjectService;
+    this.ordsProbeService = ordsProbeService;
+    this.toolRegistry = toolRegistry;
+    this.aiClient = aiClient;
+    this.objectMapper = objectMapper;
+  }
+
+  public McpReasoningResult reason(McpReasoningCommand command) {
+    ProtectedObject object = protectedObjectService.assertEnabled(command.objectId());
+    McpToolView tool = toolRegistry.toolFor(object);
+    ProbeResult probeResult = ordsProbeService.runProbe(new ProbeCommand(
+        command.objectId(),
+        command.bearerToken(),
+        normalizeLimit(command.limit())
+    ));
+    String evidenceJson = evidenceJson(tool, probeResult);
+    String prompt = buildPrompt(command.question(), tool, evidenceJson);
+
+    if (!aiClient.configured()) {
+      return new McpReasoningResult(
+          "AI_NOT_CONFIGURED",
+          tool.name(),
+          fallbackAnswer(command.question(), probeResult),
+          prompt,
+          evidenceJson,
+          probeResult
+      );
+    }
+
+    try {
+      String answer = aiClient.chat(systemPrompt(), prompt);
+      return new McpReasoningResult("SUCCESS", tool.name(), answer, prompt, evidenceJson, probeResult);
+    } catch (Exception e) {
+      return new McpReasoningResult(
+          "AI_CALL_FAILED",
+          tool.name(),
+          "AI 호출은 실패했지만 ORDS 도구 실행 증거는 수집했습니다. 상세: " + e.getMessage(),
+          prompt,
+          evidenceJson,
+          probeResult
+      );
+    }
+  }
+
+  private int normalizeLimit(int limit) {
+    if (limit < 1) {
+      return 50;
+    }
+    return Math.min(limit, 500);
+  }
+
+  private String evidenceJson(McpToolView tool, ProbeResult result) {
+    try {
+      ObjectNode evidence = objectMapper.createObjectNode();
+      evidence.put("toolName", tool.name());
+      evidence.put("objectId", tool.objectId());
+      evidence.put("displayName", tool.displayName());
+      evidence.put("ordsPath", tool.ordsPath());
+      evidence.put("status", result.status().name());
+      evidence.put("rowCount", result.rowCount());
+      evidence.set("columns", objectMapper.valueToTree(result.columns()));
+      evidence.set("maskedColumns", objectMapper.valueToTree(result.maskedColumns()));
+      evidence.set("rows", objectMapper.valueToTree(limitedRows(result)));
+      evidence.put("errorCode", result.errorCode());
+      evidence.put("errorMessage", result.errorMessage());
+      evidence.put("requestHeaders", result.requestHeaders());
+      evidence.put("requestPayload", result.requestPayload());
+      evidence.put("responseHeaders", result.responseHeaders());
+      evidence.put("responseBody", result.responseBody());
+      return objectMapper.writerWithDefaultPrettyPrinter().writeValueAsString(evidence);
+    } catch (Exception e) {
+      throw new AppException("MCP evidence JSON 생성 실패: " + e.getMessage());
+    }
+  }
+
+  private List<?> limitedRows(ProbeResult result) {
+    if (result.rows().size() <= MAX_EVIDENCE_ROWS) {
+      return result.rows();
+    }
+    return result.rows().subList(0, MAX_EVIDENCE_ROWS);
+  }
+
+  private String buildPrompt(String question, McpToolView tool, String evidenceJson) {
+    String normalizedQuestion = question == null || question.isBlank()
+        ? "이 ORDS/VPD 검증 결과를 권한 관점에서 요약해줘."
+        : question.trim();
+    return """
+        질문:
+        %s
+
+        사용한 도구:
+        - name: %s
+        - object: %s
+        - ordsPath: %s
+
+        도구 실행 증거(JSON):
+        %s
+
+        답변 요구사항:
+        - 한국어로 답변한다.
+        - 증거 JSON에 없는 데이터는 추측하지 않는다.
+        - VPD 행 필터, 컬럼 NULL 처리, ORDS 오류 여부를 구분한다.
+        - 운영자가 다음에 확인할 액션이 있으면 짧게 제시한다.
+        """.formatted(normalizedQuestion, tool.name(), tool.displayName(), tool.ordsPath(), evidenceJson);
+  }
+
+  private String systemPrompt() {
+    return """
+        당신은 Oracle ADB VPD/Redaction/ORDS 권한 검증 보조자입니다.
+        백오피스가 제공한 도구 실행 증거만 근거로 판단하고, 토큰 원문이나 비밀 값을 재출력하지 마세요.
+        """;
+  }
+
+  private String fallbackAnswer(String question, ProbeResult result) {
+    if (result.status() == ProbeStatus.SUCCESS) {
+      return "AI base URL/API Key 설정이 없어 모델 호출은 건너뛰었습니다. ORDS 호출은 성공했고 "
+          + result.rowCount() + "건이 반환되었습니다. 질문: " + safeQuestion(question);
+    }
+    return "AI base URL/API Key 설정이 없어 모델 호출은 건너뛰었습니다. ORDS 도구 실행 상태는 "
+        + result.status().name() + "입니다. 오류: "
+        + (result.errorMessage() == null ? "없음" : result.errorMessage());
+  }
+
+  private String safeQuestion(String question) {
+    if (question == null || question.isBlank()) {
+      return "기본 요약";
+    }
+    return question.trim();
+  }
+}