Testing MCP

A universal AI-powered testing server built on the Model Context Protocol (MCP).

Allows AI agents to inspect, execute, test, monitor, debug, and report on software projects.

Quick Start

# Install
pip install -e ".[dev]"

# Install Playwright browsers (for UI testing)
playwright install chromium

# Optional security tools
pip install bandit semgrep trivy

# Start server (SSE mode)
testing-mcp serve

# Or stdio mode (for Claude Desktop, Cursor, etc.)
testing-mcp stdio

Related MCP server: Gravitas-Core-MCP

MCP Client Configuration

Add to your MCP client config (claude_desktop_config.json, ~/.config/opencode/opencode.json, etc.):

{
  "mcpServers": {
    "testing-mcp": {
      "command": "testing-mcp",
      "args": ["stdio"]
    }
  }
}

Or copy testing-mcp.json to your project root for auto-discovery.

Configuration

Create a testing-mcp.toml file in your project root:

[server]
host = "127.0.0.1"
port = 8080

MCP Tools (62 total)

Core

Testing

Security

Browser Automation (Stealth)

Visual

API & Protocol

Platform

Profiling

Security Scanning Details

SAST (30 built-in rules)

Detects 15+ vulnerability classes:

• SQL Injection — string concatenation, f-string, raw queries

• XSS — innerHTML, dangerouslySetInnerHTML, unsafe template rendering

• Command Injection — os.system, subprocess(shell=True), exec/eval, popen

• SSTI — Template injection (Jinja2, Flask render_template_string)

• Path Traversal — open/Path with user-controlled paths, unsafe joins

• SSRF — User-controlled URLs in requests/httpx/fetch/urlopen

• CSRF — csrf_exempt decorators, disabled Django CSRF middleware

• IDOR — Insecure direct object references in params

• LDAP Injection — Unsanitized LDAP search filters

• XXE — XML parser with external entity expansion enabled

• Deserialization — pickle.loads, yaml.load, marshal.load

• Hardcoded Secrets — passwords, API keys, tokens in source code

• Weak Crypto — MD5, SHA1, DES, RC4, ECB mode

• Weak CORS — Access-Control-Allow-Origin: * with credentials

• Prototype Pollution — Unsafe object merge/assign

• Mass Assignment — **kwargs or update() with request data

• Race Condition — TOCTOU patterns in file operations

• File Upload — Unrestricted file upload without validation

• ReDoS — Unsafe regex patterns with user input

• Info Disclosure — Stack traces, debug endpoints exposed

• Debug Mode — Flask debug mode enabled in production

Secret Scanner (25+ patterns)

Scans for: AWS keys, GitHub tokens, GitLab tokens, Slack tokens, Discord tokens, JWT tokens, Stripe keys, SSH private keys, PGP private keys, Google API keys, Azure keys, GCP keys, Telegram tokens, Twilio credentials, Heroku API keys, npm tokens, npmrc auth tokens, database connection strings (MySQL, PostgreSQL, MongoDB, Redis), and generic bearer/token patterns

HTTP Header Scanner (15 checks)

Validates: Strict-Transport-Security (incl. weak max-age), Content-Security-Policy (incl. unsafe script-src 'unsafe-inline'), X-Frame-Options, X-Content-Type-Options, Referrer-Policy, Permissions-Policy, Cross-Origin-Embedder-Policy, Cross-Origin-Opener-Policy, Cross-Origin-Resource-Policy, Server disclosure, X-Powered-By disclosure

TLS/SSL Checker

Validates certificates, expiry dates, SAN entries, protocol versions

DAST (Dynamic Analysis)

• Technology fingerprinting (server headers, cookies, framework patterns)

• Endpoint discovery (JS source mapping, common API paths)

• SQL injection detection (error-based, time-based)

• XSS detection (reflected payloads)

• Authentication-aware scanning

Dependency Scanner

Scans lockfiles via OSV.dev API:

• requirements.txt, Pipfile.lock, poetry.lock

• package-lock.json, Cargo.lock, Gemfile.lock, go.sum

Configuration Scanner

Docker Checks (10): Running as root, latest tag used, ADD vs COPY, secrets in build args, exposed ports, HEALTHCHECK missing, no .dockerignore, vulnerable base images, shell in ENTRYPOINT, multi-stage build missing K8s Checks (10): Privileged containers, hostNetwork, runAsNonRoot, readOnlyRootFilesystem, allowPrivilegeEscalation, resource limits, seccomp, securityContext, hostPID/IPC, ServiceAccount token mount CI/CD Checks (5): Hardcoded secrets, unpinned actions, script injection, missing code checkout, missing artifact upload

Web Crawler

Discovers endpoints and files via:

• HTML source patterns (forms, links, scripts, API endpoints)

• JS source scanning (XHR/fetch URLs, route patterns)

• Common files (robots.txt, sitemap.xml, security.txt, .well-known/)

Reporter

Unified report builder that aggregates all scan results with:

• Deduplication of findings across scanners

• Risk scoring (NONE → INFO → LOW → MEDIUM → HIGH → CRITICAL)

• Text and JSON output formats

Project Structure

testing_mcp/
├── analyzers/      # Project detection (language, framework, deps)
├── api/            # API testing (httpx, OpenAPI, GraphQL, WebSocket, gRPC)
├── browser/        # Stealth browser automation (anti-bot fingerprinting, Cloudflare bypass)
├── database/       # DB validation (migrations, constraints)
├── fix/            # Remediation tools (gitignore, Docker, CI, migrations)
├── generators/     # AI test generation
├── performance/    # Benchmarking (Locust, load/stress testing, profiling)
├── plugins/        # Plugin framework (Language, Framework, Reporter)
├── reporters/      # Report generation (Markdown/HTML/JSON)
├── runners/        # Test executors + analysis (Python, Java, JS, Go, Rust, mobile, game)
├── security/       # SAST (30 rules), secrets (25+ patterns), headers, TLS, DAST,
│                   # dependency scanning (OSV.dev), config (Docker/K8s/CI),
│                   # web crawling, reporter
├── server/         # MCP server + tool registration
├── tests/          # Test suite (179+ tests covering all modules)
├── ui/             # Playwright UI automation + accessibility + visual regression
└── utils/          # Configuration, utilities

Running Tests

pip install -e ".[dev]"
pytest testing_mcp/tests/ -v

Docker

docker compose up

License

LGPL v2.1