MCP Discovery & Security Analysis Tool

A comprehensive command-line tool that scans .NET assemblies to discover MCP (Model Context Protocol) server capabilities and analyze them for security vulnerabilities. It performs deep reflection-based analysis to identify tools, resources, and prompts, while detecting potential security risks including prompt injection, tool poisoning, toxic flows, hardcoded secrets, and missing security controls.

What This Repository Contains

This repository includes:

• mcp-discover-dotnet - The main discovery tool that scans assemblies for MCP metadata
• mcp-discover.tests - A working test MCP server with examples of all capability types

What the Tool Does

The mcp-discover tool provides comprehensive analysis of .NET MCP server assemblies:

MCP Discovery Capabilities

1. Scanning all assemblies in a specified directory
2. Discovering classes and methods decorated with MCP attributes:
   • [McpServerToolType] / [McpServerTool] - Executable functions/tools
   • [McpServerResourceType] / [McpServerResource] - Static content/resources
   • [McpServerPromptType] / [McpServerPrompt] - AI prompt templates
3. Extracting metadata including names, descriptions, and audience targeting
4. Generating structured output in multiple formats (JSON, SARIF, CSV, Markdown)

Security Analysis Capabilities

1. Vulnerability Detection across 4 categories (20+ heuristics)
2. Secrets Detection for hardcoded credentials (API keys, passwords, tokens)
3. Input Validation Tracking to reduce false positives
4. Audit Logging Analysis to identify missing security logging
5. CWE Mapping for industry-standard vulnerability classification
6. Remediation Guidance with code examples and documentation links
7. CI/CD Integration with configurable severity thresholds and exit codes
8. File-Based Filtering to ignore specific files for certain vulnerability categories

Output Formats

The tool supports multiple output formats:

JSON (default) - mcp-metadata.json:

• Assembly information
• Discovered classes with MCP attributes
• Methods/members with their metadata (name, title, description, audiences)
• Security analysis findings with CWE mappings and remediation guidance
• Timestamp of generation

Markdown - mcp-metadata.md (with --markdown flag):

• Human-readable summary
• Tables organized by capability type (Tools, Resources, Prompts)
• Security findings grouped by severity (Critical, High, Medium, Low)
• Assembly details and statistics

SARIF 2.1.0 - security-report.sarif (with --format sarif):

• Static Analysis Results Interchange Format
• Industry-standard vulnerability reporting
• Integration with GitHub Code Scanning and other SAST tools
• CWE mappings, severity levels, and remediation guidance

CSV - security-report.csv (with --format csv):

• Excel-compatible tabular format
• Columns: Category, Severity, CWE, Title, Description, Location, Recommendation, Evidence
• Easy filtering and sorting for security review

Installation & Usage

Prerequisites

• .NET 10.0 SDK or later

Building from Source

# Clone the repository
git clone <repository-url>
cd mcp.discovery.tool

# Build the discovery tool
cd mcp-discover-dotnet
dotnet build -c Release

# The executable will be at:
# ./bin/Release/net10.0/mcp-discover-dotnet.exe (Windows)
# ./bin/Release/net10.0/mcp-discover-dotnet (Linux/macOS)

Using as a .NET Tool (Recommended)

The tool can be installed as a global .NET tool as its available on nuget.org:

# Install globally
dotnet tool install --global --add-source ./bin/Release mcp-discover-dotnet

# Use from anywhere
mcp-discover-dotnet <input-directory> <output-directory>

Direct Usage

# Run directly
mcp-discover-dotnet.exe <input-directory> <output-directory>

# Or with dotnet
dotnet run --project mcp-discover-dotnet -- <input-directory> <output-directory>

Command-Line Syntax

mcp-discover-dotnet <input-directory> <output-directory> [options]

Arguments:
  input-directory   Directory containing .dll assemblies to scan
  output-directory  Directory where output files will be written

Basic Options:
  -h, --help         Show help information
  -m, --markdown     Generate markdown report alongside JSON
  -o, --omit-path    Omit base path from assembly paths in output
  -n, --no-timestamp Omit timestamp from output (for version control)
  -s, --security     Perform security vulnerability analysis

Security Analysis Options:
  --config <path>           Load configuration from JSON file
  --format <format>         Export format: json (default), sarif, csv, or all
  --min-severity <level>    Minimum severity to report: Low, Medium, High, Critical
  --exclude <categories>    Comma-separated categories to exclude
  --fail-on-critical        Exit with code 1 if critical vulnerabilities found
  --fail-on-high            Exit with code 1 if high+ vulnerabilities found
  --verbose                 Display detailed security findings with CWE mappings

Examples

# Basic MCP discovery
mcp-discover-dotnet ./bin/Release/net10.0 ./metadata

# Generate both JSON and markdown reports
mcp-discover-dotnet ./bin/Release/net10.0 ./metadata --markdown

# Perform security vulnerability scan
mcp-discover-dotnet ./bin/Release/net10.0 ./metadata --security

# Security scan with SARIF export for CI/CD integration
mcp-discover-dotnet ./bin/Release/net10.0 ./metadata --security --format sarif

# Export in all formats (JSON, SARIF, CSV)
mcp-discover-dotnet ./bin/Release/net10.0 ./metadata --security --format all

# Filter to show only High and Critical findings
mcp-discover-dotnet ./bin/Release/net10.0 ./metadata --security --min-severity High

# Exclude specific vulnerability categories
mcp-discover-dotnet ./bin/Release/net10.0 ./metadata --security --exclude PromptInjection,ToxicFlow

# CI/CD integration: fail build if critical vulnerabilities found
mcp-discover-dotnet ./bin/Release/net10.0 ./metadata --security --fail-on-critical

# Comprehensive analysis with verbose output
mcp-discover-dotnet ./bin/Release/net10.0 ./metadata --markdown --security --format all --verbose

# Use configuration file for custom patterns and thresholds
mcp-discover-dotnet ./bin/Release/net10.0 ./metadata --security --config security-config.json

# Use relative paths in output instead of full paths
mcp-discover-dotnet ./bin/Release/net10.0 ./metadata --omit-path --security

Security Vulnerability Detection

The tool performs comprehensive security analysis of MCP servers with pattern-based heuristics and enhanced detection:

Vulnerability Categories

1. Prompt Injection Attacks 💉 (CWE-74)

   • Detects prompts that may accept unsanitized user input
   • Identifies string concatenation patterns in prompt descriptions
   • Flags missing input validation
   • Tracks methods with validation to reduce false positives

2. Tool Poisoning Attacks ☠️ (CWE-494)

   • Identifies dangerous operations (file system, command execution, database access)
   • Detects file system operations vulnerable to path traversal
   • Flags database operations that may be vulnerable to SQL injection
   • Warns about external API calls susceptible to SSRF
   • Reduces severity when validation methods are detected

3. Toxic Flow Issues ⚡ (CWE-693)

   • Detects async operations without timeout configuration
   • Identifies expensive operations without rate limiting
   • Flags resource exhaustion risks

4. General Security Issues 🛡️ (CWE-312, CWE-778)

   • Secrets Detection: Identifies hardcoded credentials (API keys, passwords, tokens, bearer tokens, AWS secrets, private keys)
   • Audit Logging: Detects missing audit logging in sensitive operations
   • Identifies missing authorization controls on sensitive operations
   • Detects lack of audience restrictions
   • Flags tools accessing external resources without validation

Enhanced Detection Features

• Input Validation Tracking: Analyzes methods for validation patterns (validate, sanitize, check, verify, whitelist) to reduce false positives
• CWE Mapping: Maps all findings to Common Weakness Enumeration IDs for industry-standard classification
• Remediation Guidance: Provides actionable code examples and OWASP documentation links
• Secrets Patterns: 13 patterns for detecting hardcoded credentials
• Validation Patterns: 7 patterns for identifying input validation
• Logging Patterns: 5 patterns for detecting audit logging

Severity Levels

• Critical: Immediate security risk requiring urgent attention
• High: Significant security concern that should be addressed soon
• Medium: Potential security issue worth reviewing
• Low: Minor security consideration or best practice improvement

Security Scan Output

When --security is used, the JSON output includes a comprehensive SecurityAnalysis section:

{
  "SecurityAnalysis": {
    "TotalFindings": 12,
    "CriticalCount": 2,
    "HighCount": 5,
    "MediumCount": 3,
    "LowCount": 2,
    "Findings": [
      {
        "Category": "GeneralSecurity",
        "Severity": "Critical",
        "CWE": "CWE-312",
        "Title": "Hardcoded Credentials Detected",
        "Description": "Method contains hardcoded secrets or credentials",
        "Location": "AuthService.GetApiKey",
        "Recommendation": "Use secure secret management (Azure Key Vault, AWS Secrets Manager)",
        "Evidence": "Pattern 'api_key' detected",
        "CodeExample": "// Use environment variables or secure vaults\nvar apiKey = Environment.GetEnvironmentVariable(\"API_KEY\");",
        "DocumentationLink": "https://owasp.org/www-community/vulnerabilities/Use_of_hard-coded_credentials"
      },
      {
        "Category": "ToolPoisoning",
        "Severity": "High",
        "CWE": "CWE-494",
        "Title": "Potentially Dangerous Tool Operation",
        "Description": "Tool contains dangerous operation pattern",
        "Location": "FileTools.DeleteFile",
        "Recommendation": "Implement strict input validation, path sanitization, and authorization checks",
        "Evidence": "Pattern 'Delete' detected",
        "CodeExample": "// Validate and sanitize file paths\nvar safePath = Path.GetFullPath(userInput);\nif (!safePath.StartsWith(allowedDirectory)) throw new SecurityException();",
        "DocumentationLink": "https://owasp.org/www-community/attacks/Path_Traversal"
      }
    ]
  }
}

SARIF 2.1.0 Export

Use --format sarif for industry-standard Static Analysis Results Interchange Format:

mcp-discover-dotnet ./bin/Release/net10.0 ./output --security --format sarif

This generates security-report.sarif compatible with:

• GitHub Code Scanning
• Azure DevOps Security Analysis
• SonarQube
• Other SAST tools

CSV Export

Use --format csv for Excel-compatible tabular export:

mcp-discover-dotnet ./bin/Release/net10.0 ./output --security --format csv

Columns: Category, Severity, CWE, Title, Description, Location, Recommendation, Evidence

Markdown Report

With --markdown --security, the report includes a detailed security section with findings grouped by severity and color-coded output.

Configuration File

Create a JSON configuration file to customize security scanning behavior:

{
  "MinimumSeverity": null,
  "ExcludeCategories": [],
  "Patterns": {
    "SecretsPatterns": [
      "api_key",
      "password",
      "custom_secret"
    ],
    "ValidationPatterns": [
      "validate",
      "sanitize"
    ],
    "LoggingPatterns": [
      "log",
      "audit"
    ]
  },
  "Thresholds": {
    "CriticalThreshold": 0,
    "HighThreshold": 5
  },
  "Suppressions": [
    {
      "Location": "TestClass.TestMethod",
      "Reason": "False positive - method is for testing only"
    }
  ],
  "IgnoreFiles": {
    "PromptInjection": [
      "**/Tests/**",
      "**/TestData/**"
    ],
    "GeneralSecurity": [
      "**/Mock*.dll"
    ]
  }
}

Configuration Options

MinimumSeverity (null | "Low" | "Medium" | "High" | "Critical")

• Filter findings below this severity level
• null shows all findings

ExcludeCategories (array of strings)

• Skip entire vulnerability categories
• Valid values: "PromptInjection", "ToolPoisoning", "ToxicFlow", "GeneralSecurity"

Patterns (object)

• SecretsPatterns: Additional patterns for detecting hardcoded credentials
• ValidationPatterns: Patterns indicating input validation (reduces false positives)
• LoggingPatterns: Patterns for detecting audit logging

Thresholds (object)

• CriticalThreshold: Maximum critical findings before CI/CD failure (with --fail-on-critical)
• HighThreshold: Maximum high+ findings before CI/CD failure (with --fail-on-high)

Suppressions (array of objects)

• Location: Method or class to suppress (e.g., "MyClass.MyMethod")
• Reason: Documentation of why this finding is suppressed

IgnoreFiles (object)

• Key: Vulnerability category name
• Value: Array of glob patterns for files to ignore
• Supports wildcards: ** (any directory), * (any characters)
• Examples:
  • "**/Tests/**" - Ignore all files in any Tests directory
  • "**/Mock*.dll" - Ignore DLLs starting with "Mock"
  • "**/bin/Debug/**" - Ignore debug builds

Using Configuration Files

# Basic usage with config
mcp-discover-dotnet ./bin/Release/net10.0 ./output --security --config security-config.json

# Override config severity with command-line flag
mcp-discover-dotnet ./bin/Release/net10.0 ./output --security --config security-config.json --min-severity High

# Combine with other options
mcp-discover-dotnet ./bin/Release/net10.0 ./output --security --config security-config.json --format sarif --verbose

CI/CD Integration

Exit Codes

The tool uses exit codes for build automation:

• Exit 0: Success (no failures, or thresholds not exceeded)
• Exit 1: Failure (thresholds exceeded when using --fail-on-* flags)

GitHub Actions Example

name: MCP Security Scan

on: [push, pull_request]

jobs:
  security:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v3
      
      - name: Setup .NET
        uses: actions/setup-dotnet@v3
        with:
          dotnet-version: '10.0.x'
      
      - name: Install MCP Discovery Tool
        run: dotnet tool install --global mcp-discover-dotnet
      
      - name: Build Application
        run: dotnet build -c Release
      
      - name: Run Security Scan
        run: |
          mcp-discover-dotnet ./bin/Release/net10.0 ./security-report \
            --security \
            --format sarif \
            --fail-on-critical \
            --config security-config.json
      
      - name: Upload SARIF to GitHub Security
        uses: github/codeql-action/upload-sarif@v2
        if: always()
        with:
          sarif_file: security-report/security-report.sarif

Azure DevOps Pipeline Example

trigger:
  - main

pool:
  vmImage: 'ubuntu-latest'

steps:
- task: UseDotNet@2
  inputs:
    version: '10.0.x'

- script: dotnet tool install --global mcp-discover-dotnet
  displayName: 'Install MCP Discovery Tool'

- script: dotnet build -c Release
  displayName: 'Build Application'

- script: |
    mcp-discover-dotnet ./bin/Release/net10.0 ./security-report \
      --security \
      --format all \
      --fail-on-high \
      --verbose
  displayName: 'Security Scan'

- task: PublishBuildArtifacts@1
  condition: always()
  inputs:
    pathToPublish: 'security-report'
    artifactName: 'security-analysis'

Threshold-Based Gating

# Fail build if ANY critical vulnerabilities found
mcp-discover-dotnet ./bin ./output --security --fail-on-critical

# Fail build if ANY high or critical vulnerabilities found  
mcp-discover-dotnet ./bin ./output --security --fail-on-high

# Use config file to set numeric thresholds
# security-config.json: { "Thresholds": { "CriticalThreshold": 0, "HighThreshold": 5 } }
mcp-discover-dotnet ./bin ./output --security --config security-config.json --fail-on-high

Quick Start with Test Server

The repository includes a complete test server that demonstrates all MCP capability types:

# Navigate to the test server
cd mcp-discover.tests

# Build (automatically runs discovery tool)
dotnet build -c Release

# Run the server
.\run.ps1

# View generated metadata
.\view-metadata.ps1

See mcp-discover.tests/README.md for detailed documentation.

MCP Attributes Reference

To make your .NET assemblies discoverable, use these attributes:

Class-Level Attributes

[McpServerToolType("group-name", "Display Title")]
public class MyTools { }

[McpServerResourceType("group-name", "Display Title")]
public class MyResources { }

[McpServerPromptType("group-name", "Display Title")]
public class MyPrompts { }

Method-Level Attributes

[McpServerTool("tool-name", "Display Title")]
[Description("Detailed description of what this tool does")]
[McpAudience("developers", "qa")]
public static int MyTool(int param) { return param; }

[McpServerResource("resource-name", "Display Title")]
public string GetResource() { return "content"; }

[McpServerPrompt("prompt-name", "Display Title")]
public string GetPrompt() { return "prompt template"; }

Metadata Attributes

[Description("Detailed description")]
[McpAudience("audience1", "audience2")]

Example Output

{
  "GeneratedAtUtc": "2025-11-22T10:30:00Z",
  "Assemblies": [
    {
      "AssemblyPath": "C:\\path\\to\\server.dll",
      "Classes": [
        {
          "TypeName": "MyApp.CalculatorTools",
          "Kind": "ToolType",
          "Description": "Math operations",
          "Audiences": ["developers"],
          "Members": [
            {
              "MethodName": "Add",
              "Kind": "Tool",
              "Name": "add",
              "Title": "Add Numbers",
              "Description": "Adds two integers",
              "Audiences": ["developers"]
            }
          ]
        }
      ]
    }
  ]
}

Integration with Build Process

You can integrate the discovery tool into your MSBuild process:

<Target Name="RunMcpDiscovery" AfterTargets="Build">
  <PropertyGroup>
    <McpDiscoverTool>path\to\mcp-discover-dotnet.exe</McpDiscoverTool>
    <InputDir>$(MSBuildProjectDirectory)\$(OutputPath)</InputDir>
    <OutputDir>$(MSBuildProjectDirectory)\$(OutputPath)mcp-metadata</OutputDir>
  </PropertyGroup>
  
  <Exec Command="&quot;$(McpDiscoverTool)&quot; &quot;$(InputDir)&quot; &quot;$(OutputDir)&quot;" />
</Target>

See mcp-discover.tests/mcp-discover.tests.csproj for a working example.

Repository Structure

mcp.discovery.tool/
├── mcp-discover-dotnet/       # Discovery tool source code
│   ├── Program.cs             # Main scanning logic
│   └── mcp-discover-dotnet.csproj    # Tool project file
│
├── mcp-discover.tests/        # Test MCP server
│   ├── SampleServer.cs        # Example server implementation
│   ├── McpAttributes.cs       # Attribute definitions
│   ├── run.ps1                # Server runner script
│   ├── view-metadata.ps1      # Metadata viewer script
│   └── README.md              # Test server documentation
│
├── mcp.discovery.tool.sln     # Solution file
└── README.md                  # This file

Use Cases

MCP Development

• Automated Capability Discovery: Generate capability metadata during build
• Documentation Generation: Extract and document available tools/resources/prompts
• Validation: Verify that MCP attributes are correctly applied
• Dynamic Discovery: Discover capabilities in MCP server assemblies at runtime

Security & Compliance

• Vulnerability Detection: Identify security risks in MCP servers before deployment
• CI/CD Security Gating: Block builds with critical vulnerabilities
• SARIF Integration: Feed findings into GitHub Code Scanning, Azure DevOps, SonarQube
• Compliance Reporting: Generate CSV/Excel reports for security audits
• CWE Mapping: Track vulnerabilities using industry-standard classification

DevSecOps

• Shift-Left Security: Catch vulnerabilities early in development
• Automated Scanning: Integrate into build pipelines for continuous security analysis
• False Positive Reduction: Input validation tracking reduces noise
• Secrets Detection: Prevent hardcoded credentials from reaching production
• Security Metrics: Track Critical/High/Medium/Low findings over time

Testing & Quality Assurance

• Security Testing: Validate MCP server implementations for security best practices
• Regression Detection: Track security findings across versions
• Suppression Management: Document and manage known false positives

How It Works

MCP Discovery

1. Assembly Loading: Uses isolated AssemblyLoadContext to safely load assemblies
2. Reflection: Scans types and methods for MCP attributes
3. Metadata Extraction: Reads attribute constructor arguments and properties
4. Output Generation: Serializes discovered metadata to structured formats

Security Analysis

1. Pattern Matching: Applies 20+ heuristics across 4 vulnerability categories
2. Validation Tracking: Identifies methods with input validation to reduce false positives
3. Secrets Detection: Scans for 13 patterns of hardcoded credentials
4. Audit Logging: Detects missing security logging in sensitive operations
5. CWE Mapping: Maps findings to Common Weakness Enumeration IDs
6. Remediation: Provides code examples and OWASP documentation links
7. Filtering: Applies severity, category, and file-based filtering
8. Export: Generates JSON, SARIF 2.1.0, CSV, and Markdown reports
9. CI/CD: Returns exit codes based on configurable severity thresholds

Requirements

• .NET 10.0 SDK or later
• Windows, Linux, or macOS
• Assemblies must be compiled .NET assemblies (.dll files)

Contributing

Contributions are welcome! Areas for improvement:

Features

• Support for additional metadata attributes
• More vulnerability detection heuristics
• Custom rule engine for security analysis
• Integration with additional SAST tools

Performance

• Performance optimizations for large assembly sets
• Parallel processing for multi-assembly analysis
• Incremental scanning for unchanged files

Export & Integration

• Additional output formats (XML, HTML reports)
• Integration examples for other CI/CD systems (GitLab CI, Jenkins)
• VS Code extension for inline security warnings
• NuGet package for programmatic API access

Author

echapmanFromBunnings

Related Resources

• Model Context Protocol (MCP) - Learn about the MCP standard
• Test Server Documentation - Detailed example implementation
• Verification Guide - Testing and validation steps

---

Need Help? Check the test server in mcp-discover.tests/ for a complete working example!