Skip to content

APILeak v1.0.0

APILeak Logo

Enterprise-grade API security testing tool with full OWASP API Security Top 10 2023 coverage, JWT attack testing, and a built-in MCP server that lets AI agents like Claude or Kiro run security scans via natural language.

License: MIT Python 3.11+ OWASP API Security MCP

What is APILeak?

APILeak is a CLI and MCP-based tool for automated penetration testing of REST APIs. It combines:

  • 12 OWASP testing modules covering the full API Security Top 10 2023 + HPP + spec-driven testing
  • JWT attack suite — 8 attack types including none algorithm, weak secret brute-force, algorithm confusion, and key injection
  • Endpoint and parameter discovery — directory fuzzing, parameter fuzzing, OpenAPI-aware scanning
  • Cloud gateway audits — AWS API Gateway, Google Cloud API Gateway, Azure APIM
  • MCP server — exposes all of the above as tools for AI agents (Claude Desktop, Kiro, custom agents)
  • AI triage and deduplication — optional LLM-assisted finding prioritisation (Anthropic, OpenAI, Bedrock)

5-Minute Quick Start

pip install 'apileaks[mcp]'
apileaks scan --target https://api.example.com

# Pull or build
docker build -f Dockerfile.mcp -t apileak-mcp:latest .

# Run a scan
docker run --rm apileak-mcp:latest \
  python apileaks.py scan --target https://api.example.com

git clone https://github.com/MoisesTapia/apileaks.git
cd apileaks
pip install -e '.[mcp,ai]'
apileaks scan --target https://api.example.com

Start the MCP Server

Give your AI agent full access to APILeak:

Add to your Claude Desktop config (claude_desktop_config.json): 

{
  "mcpServers": {
    "apileaks": {
      "command": "apileaks",
      "args": ["mcp-server", "--transport", "stdio"]
    }
  }
}
Restart Claude Desktop. APILeak tools appear in the tools panel.

export APILEAKS_MCP_AUTH_TOKEN=$(python -c "import secrets; print(secrets.token_urlsafe(32))")
apileaks mcp-server --transport http --port 8080

echo "APILEAKS_MCP_AUTH_TOKEN=changeme" > .env
docker compose --profile mcp up -d
# Server available at http://localhost:8080/sse

Once connected, you can ask Claude:

"Run a full OWASP scan against https://staging.mycompany.com/api using Bearer token eyJ..."

And Claude will invoke apileak.scan with the appropriate parameters, stream progress, and return a structured finding report.

What Can the AI Agent Do?

Capability Tool Notes
Full OWASP scan apileak.scan All 12 modules + discovery
BOLA / IDOR testing apileak.owasp.bola_testing API1:2023
JWT attacks apileak.jwt.attack_test 8 attack types
SSRF testing apileak.owasp.ssrf_testing Cloud metadata, blind SSRF
Auth testing apileak.owasp.auth_testing API2:2023
Mass assignment apileak.owasp.property_level_auth API3:2023
Endpoint discovery apileak.discovery.dir Wordlist fuzzing
Parameter discovery apileak.discovery.par Hidden parameters
Cloud audit apileak.cloud_audit AWS, GCP, Azure ⚠️ Destructive
+ 6 more OWASP modules … API4–API10 + HPP + Spec

Destructive tools (those that modify state) require a ConfirmationToken — the AI cannot execute them without explicit human confirmation.

Documentation Map

  • Getting Started

    Install APILeak, set up your environment, run your first scan.

  • MCP Server Guide

    Connect Claude Desktop, Kiro, or any MCP client to APILeak in minutes.

  • Docker Deployment

    Run the MCP server in containers with docker compose.

  • OWASP Modules

    Deep dives into each of the 12 OWASP testing modules.

  • JWT Attack Testing

    Comprehensive JWT attack suite — 8 attack types explained.

  • AI Providers

    Configure Anthropic, OpenAI, or Amazon Bedrock for AI triage.

  • MCP Tools Reference

    Complete JSON Schema and examples for every MCP tool.

  • MCP Security

    OWASP MCP Top 10 controls, self-audit results, security policies.

Command Overview

Command Purpose
apileaks scan Orchestrated scan — discovery + all (or selected) OWASP modules
apileaks owasp <module> Run a single OWASP module in isolation
apileaks dir Directory/endpoint fuzzing
apileaks par Hidden parameter discovery
apileaks jwt <subcommand> JWT attack and utility toolkit
apileaks mcp-server Start the MCP server (stdio or HTTP)
apileaks mcp-audit Audit a third-party MCP server for security issues
apileaks cloud-audit Cloud API gateway security audit
apileaks replay <report> Re-issue requests from a prior scan report

Security Philosophy

APILeak is built with the same security controls it tests for in others:

  • No AI-initiated destructive actions — AI agents must receive a human confirmation token for any destructive scan
  • Append-only audit log — every tool invocation is recorded; no handler can erase it
  • Hash-verified supply chain — MCP SDK pinned with ==; requirements-mcp.txt generated with pip-compile --generate-hashes
  • Self-audit (dogfooding) — apileaks mcp-audit runs against the MCP server itself in CI before every release
  • OWASP MCP Top 10 — all 10 controls implemented; see MCP Security docs

Contributing

See CONTRIBUTING.md for: - Development setup - Coding standards (ruff, mypy, pytest) - MCP SDK version bump policy - Pull request workflow

License

MIT License — free to use, modify, and distribute.