feat: Audit mode with FP reduction for automated MCP auditing

CLAUDE.md

@@ -14,8 +14,9 @@
 - Development mode: `npm run dev` (use `npm run dev:windows` on Windows)
 - Format code: `npm run prettier-fix`
 - Client lint: `cd client && npm run lint`
-- Run tests: `npm test` (464 passing, 100% pass rate)
+- Run tests: `npm test` (696 total, 669 passing, 27 timeout failures)
 - Run assessment tests: `npm test -- assessment` (208 assessment module tests)
+- Note: 27 test failures are SecurityAssessor timeout issues (480s limit), not logic errors
 
 ## Code Style Guidelines
 
@@ -195,6 +196,7 @@ For detailed documentation on specific features, see:
 - **Functionality Testing**: [README.md](README.md#2-optimized-progressive-complexity-testing) - Multi-scenario validation, progressive complexity
 - **Security Assessment**: [README.md](README.md#4-context-aware-security-assessment-with-zero-false-positives) - Domain-specific patterns, zero false positives
 - **Error Handling**: [README.md](README.md#assessment-categories) - MCP protocol compliance, validation quality
+- **MCP Spec Reference**: [docs/mcp_spec_11-2025.md](docs/mcp_spec_11-2025.md) - Protocol revision 2025-11-25 (latest)
 - **MCP Spec Compliance**: See PROJECT_STATUS.md timeline for latest enhancements
 - **Recent Changes**: [PROJECT_STATUS.md](PROJECT_STATUS.md#development-timeline---october-2025)
 
@@ -217,7 +219,7 @@ For detailed documentation on specific features, see:
 
 **Testing:**
 
-- `client/src/services/__tests__/` - 464 total tests (100% passing)
+- `client/src/services/__tests__/` - 696 total tests (669 passing, 27 timeout failures)
 - `client/src/services/assessment/__tests__/` - 208 assessment module tests
 
 ## Development Workflow

README.md

@@ -353,7 +353,7 @@ Our enhanced MCP Inspector includes a comprehensive assessment system that valid
    - Input validation and sanitization checks
    - Authentication/authorization testing
    - Sensitive data exposure detection
-   - Dual-mode testing: Reviewer mode (3 critical patterns) + Developer mode (all 17 patterns)
+   - Dual-mode testing: Reviewer mode (3 critical patterns) + Developer mode (all 13 patterns)
 
 5. **Usability**
    - Tool naming consistency analysis

cli/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@bryan-thompson/inspector-assessment-cli",
-  "version": "1.7.1",
+  "version": "1.8.0",
   "description": "CLI for the Enhanced MCP Inspector with assessment capabilities",
   "license": "MIT",
   "author": "Bryan Thompson <bryan@triepod.ai>",

cli/src/assess-full.ts

@@ -48,6 +48,7 @@ interface AssessmentOptions {
   sourceCodePath?: string;
   claudeEnabled?: boolean;
   fullAssessment?: boolean;
+  auditMode?: boolean;
   verbose?: boolean;
   jsonOnly?: boolean;
   helpRequested?: boolean;
@@ -286,7 +287,22 @@ function buildConfig(options: AssessmentOptions): AssessmentConfiguration {
     testTimeout: 30000,
   };
 
-  if (options.fullAssessment !== false) {
+  if (options.auditMode) {
+    // Audit mode: only HIGH-value modules for automated MCP auditing
+    config.assessmentCategories = {
+      functionality: true,
+      security: true,
+      documentation: false,
+      errorHandling: true,
+      usability: false,
+      mcpSpecCompliance: true,
+      aupCompliance: false,
+      toolAnnotations: true,
+      prohibitedLibraries: false,
+      manifestValidation: false,
+      portability: false,
+    };
+  } else if (options.fullAssessment !== false) {
     config.assessmentCategories = {
       functionality: true,
       security: true,
@@ -376,6 +392,7 @@ async function runFullAssessment(
     callTool: createCallToolWrapper(client),
     config,
     sourceCodePath: options.sourceCodePath,
+    transportType: serverConfig.transport || "stdio",
     ...sourceFiles,
   };
 
@@ -400,13 +417,53 @@ function saveResults(
   serverName: string,
   results: MCPDirectoryAssessment,
   outputPath?: string,
+  transportType?: string,
 ): string {
   const defaultPath = `/tmp/inspector-full-assessment-${serverName}.json`;
   const finalPath = outputPath || defaultPath;
 
+  // Build audit summary for automated consumption
+  const securityResult = results.security as {
+    auditAnalysis?: {
+      highConfidenceVulnerabilities: string[];
+      needsReview: string[];
+      falsePositiveLikelihood: Record<string, string>;
+    };
+    vulnerabilities?: string[];
+  };
+  const functionalityResult = results.functionality as {
+    workingTools?: number;
+    totalTools?: number;
+  };
+  const mcpResult = results.mcpSpecCompliance as {
+    metrics?: { overallScore?: number };
+  };
+  const errorResult = results.errorHandling as {
+    metrics?: { mcpComplianceScore?: number };
+  };
+
+  const auditSummary = {
+    highConfidenceVulnerabilities:
+      securityResult?.auditAnalysis?.highConfidenceVulnerabilities || [],
+    needsReview: securityResult?.auditAnalysis?.needsReview || [],
+    falsePositiveLikelihood:
+      securityResult?.auditAnalysis?.falsePositiveLikelihood || {},
+    functionalTools: functionalityResult?.workingTools || 0,
+    totalTools: functionalityResult?.totalTools || 0,
+    mcpComplianceScore: errorResult?.metrics?.mcpComplianceScore || 0,
+    transportType: transportType || "unknown",
+    recommendedAction:
+      results.overallStatus === "PASS"
+        ? "APPROVE"
+        : results.overallStatus === "FAIL"
+          ? "REJECT"
+          : "REVIEW",
+  };
+
   const output = {
     timestamp: new Date().toISOString(),
     assessmentType: "full",
+    auditSummary,
     ...results,
   };
 
@@ -563,6 +620,9 @@ function parseArgs(): AssessmentOptions {
       case "--full":
         options.fullAssessment = true;
         break;
+      case "--audit-mode":
+        options.auditMode = true;
+        break;
       case "--verbose":
       case "-v":
         options.verbose = true;
@@ -617,6 +677,9 @@ Options:
   --source <path>        Source code path for deep analysis (AUP, portability, etc.)
   --claude-enabled       Enable Claude Code integration for intelligent analysis
   --full                 Enable all assessment modules (default)
+  --audit-mode           Run only high-value modules for automated MCP auditing
+                         (Functionality, Security, ErrorHandling, MCPSpecCompliance, ToolAnnotations)
+                         Reduces false positives and includes audit summary in output
   --json                 Output only JSON (no console summary)
   --verbose, -v          Enable verbose logging
   --help, -h             Show this help message
@@ -658,10 +721,16 @@ async function main() {
       displaySummary(results);
     }
 
+    // Determine transport type for audit summary
+    const serverConfig = loadServerConfig(
+      options.serverName,
+      options.serverConfigPath,
+    );
     const outputPath = saveResults(
       options.serverName,
       results,
       options.outputPath,
+      serverConfig.transport || "stdio",
     );
 
     if (options.jsonOnly) {

cli/src/assess-security.ts

@@ -237,7 +237,7 @@ async function runSecurityAssessment(
 
   const config: AssessmentConfiguration = {
     ...DEFAULT_ASSESSMENT_CONFIG,
-    securityPatternsToTest: 17,
+    securityPatternsToTest: 13,
     reviewerMode: false,
     testTimeout: 30000,
   };
@@ -247,9 +247,10 @@ async function runSecurityAssessment(
     tools,
     callTool: createCallToolWrapper(client),
     config,
+    transportType: serverConfig.transport || "stdio",
   };
 
-  console.log(`🛡️  Running security assessment with 17 attack patterns...`);
+  console.log(`🛡️  Running security assessment with 13 attack patterns...`);
   const assessor = new SecurityAssessor(config);
   const results = await assessor.assess(context);
 
@@ -386,7 +387,7 @@ function printHelp() {
   console.log(`
 Usage: mcp-assess-security [options] [server-name]
 
-Run security assessment against an MCP server with 17 attack patterns.
+Run security assessment against an MCP server with 13 attack patterns.
 
 Options:
   --server, -s <name>    Server name (required, or pass as first positional arg)
@@ -396,15 +397,14 @@ Options:
   --verbose, -v          Enable verbose logging
   --help, -h             Show this help message
 
-Attack Patterns Tested (17 total):
-  • Direct prompt injection
-  • Indirect prompt injection
-  • Instruction override
-  • Role-playing attacks
-  • Encoding bypass
-  • Multi-turn manipulation
-  • Context poisoning
-  • And more...
+Attack Patterns Tested (13 total):
+  • Command Injection        • SQL Injection
+  • Calculator Injection     • Path Traversal
+  • Type Safety              • Boundary Testing
+  • Required Fields          • MCP Error Format
+  • Timeout Handling         • Indirect Prompt Injection
+  • Unicode Bypass           • Nested Injection
+  • Package Squatting
 
 Examples:
   mcp-assess-security my-server

client/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@bryan-thompson/inspector-assessment-client",
-  "version": "1.7.1",
+  "version": "1.8.0",
   "description": "Client-side application for the Enhanced MCP Inspector with assessment capabilities",
   "license": "MIT",
   "author": "Bryan Thompson <bryan@triepod.ai>",

client/src/lib/assessmentTypes.ts

@@ -66,6 +66,9 @@ export interface SecurityTestResult {
   connectionError?: boolean; // True if test failed due to connection/server failure
   errorType?: "connection" | "server" | "protocol"; // Classify error type
   testReliability?: "completed" | "failed" | "retried"; // Test execution status
+  // Audit-mode fields for automated consumption
+  vulnerableHighConfidence?: boolean; // Only true when confidence === "high" AND vulnerable
+  toolCategory?: string; // Classified tool category (e.g., "search_retrieval", "calculator")
 }
 
 export interface CodeExample {
@@ -243,6 +246,16 @@ export interface SecurityAssessment {
   overallRiskLevel: SecurityRiskLevel;
   status: AssessmentStatus;
   explanation: string;
+  // Audit-mode: pre-computed false positive analysis
+  auditAnalysis?: {
+    highConfidenceVulnerabilities: string[];
+    needsReview: string[];
+    falsePositiveLikelihood: Record<string, "HIGH" | "MEDIUM" | "LOW">;
+    responseUniformity: Record<
+      string,
+      { uniqueResponses: number; totalTests: number }
+    >;
+  };
 }
 
 export interface DocumentationAssessment {

client/src/services/assessment/AssessmentOrchestrator.ts

@@ -66,6 +66,10 @@ export interface AssessmentContext {
   // MCPB manifest validation (optional)
   manifestJson?: ManifestJsonSchema;
   manifestRaw?: string; // Raw manifest.json content for parsing validation
+
+  // Transport type for context-aware security testing
+  // Used to skip irrelevant tests (e.g., path traversal on remote servers)
+  transportType?: "stdio" | "http" | "sse";
 }
 
 export class AssessmentOrchestrator {