Skip to main content

CyberEiche MCP-Sec 🛡️

Deterministic MCP Vulnerability Verification, Agent Fuzzing & Zero-Noise Pytest Gate

PyPI version Tests Coverage Python MCP Spec False Positives License


⚡ TL;DR — Quick Command Cheat Sheet (For the Busy Developer)

# 📦 1. Install CyberEiche MCP-Sec
pip install cybereiche-mcp-sec
# Or via uv
uv add --dev cybereiche-mcp-sec

# 🔍 2. Run Directed Fuzzing via Native Pytest
pytest --mcp-fuzz

# 🎯 3. Target a Specific MCP Server Config (Desktop / Claude / Stdio)
pytest --mcp-fuzz --mcp-target path/to/claude_desktop_config.json

# 📋 4. Export OASIS SARIF 2.1.0 Report (for GitHub Security tab & IDEs)
pytest --mcp-fuzz --mcp-sarif results.sarif

# 🔄 5. Resumable Session with Crash Salvage (Multi-day or long-running scans)
pytest --mcp-fuzz --mcp-session .cybereiche/fuzz_session.json

# 🛡️ 6. Enforce Human-in-the-Loop (HITL) Confirmation for Destructive Sinks
pytest --mcp-fuzz --mcp-hitl

# 🛠️ 7. Verify an Automated Fix Neutralizes the PoV (Closed-Loop Remediation)
pytest --mcp-verify-fix fix.patch

💥 The Real Problem We Solve

In late-2026, 87.5% of verified MCP CVEs are High or Critical severity, with over 66% leading to direct Arbitrary Code Execution (ACE) or Path Traversal via Tool Poisoning Attacks (TPAs).

Traditional security scanners give agent developers two bad choices:

  1. Noisy SAST / LLM Scanners: 80–90% false-positive noise from theoretical prompt injections and hallucinations that never actually compromise a tool.
  2. Unchecked Black-Box Attackers: Burn thousands of dollars in LLM tokens guessing blind, leak API credentials in prompt contexts, and stop at a 50-page PDF without reproducible proof.

The CyberEiche Difference:

  • 0% False-Positive Guarantee (Machine Oracle): Enforces DARPA AIxCC principles. A vulnerability is alerted if and only if an executable, reproducible test_exploit_<id>.py script is synthesized and verified.
  • Local Privacy Gateway ($SECRET_n Tokenization): Credentials, connection strings, and tokens are intercepted and tokenized before reaching LLM contexts or execution logs.
  • Closed-Loop Fix Verification: Connects verified findings directly to automated fixers (such as cybereiche-vibesec fix) and confirms that the exploit is neutralized with zero test regressions.

🏗️ Architecture & Verification Lifecycle

[1. MCP CATALOG PARSER]  ──► Deep JSON Schema & Desktop Config Discovery
                               │ *Maps dangerous sinks: FILESYSTEM, EXEC, DB, NETWORK*
                               ▼
[2. PRIVACY GATEWAY]     ──► Tokenizes credentials ($SECRET_n) & enforces scope boundaries
                               │ *Zero credential leakage to model prompts or logs*
                               ▼
[3. MULTI-TURN FUZZER]   ──► Adaptive conversational attacks (Crescendo & TAP)
                               │ *Guided multi-step steering toward dangerous sinks*
                               ▼
[4. MICRO-SANDBOX & ORACLE]► Ephemeral JSON-RPC mock server with cryptographic canaries
                               │ *Multi-encoding exfiltration detection (Raw, B64, URL, Hex)*
                               ▼
[5. MACHINE ORACLE POV]  ──► Auto-generates reproducible test_exploit_<id>.py & SARIF 2.1.0
                               │ *0% noise: Drops unverified alerts; saves session state*
                               ▼
[6. CLOSED-LOOP FIX]     ──► Re-runs PoV against patches to guarantee verified remediation

📋 Complete CLI & Pytest Options Reference

Flag Argument Description
--mcp-fuzz Flag Activates CyberEiche MCP fuzzing engine within pytest.
--mcp-target <path> Path to MCP server catalog or claude_desktop_config.json.
--mcp-sarif <path> Serializes verified vulnerabilities to standard SARIF 2.1.0 JSON.
--mcp-session <path> Checkpoints state atomically to allow resuming interrupted fuzzing runs.
--mcp-hitl Flag Prompts operator confirmation before dangerous sinks execute.
--mcp-verify-fix <path> Validates that a candidate patch neutralizes the synthesized PoV.

🛡️ The Zero-Noise Guarantee & PoV Synthesis

CyberEiche MCP-Sec outputs clear, actionable scorecards in the terminal:

           CyberEiche MCP Security Scorecard (Verified Exploits Only)                
┏━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━┳━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━┓
┃ Vulnerability ID ┃ Severity ┃ Category       ┃ Title                  ┃ Evidence   ┃
┡━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━╇━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━┩
│ VULN-TPA-001     │ CRITICAL │ tool_poisoning │ Arbitrary File Deletion│ Executed   │
│                  │          │                │                        │ delete_file│
└──────────────────┴──────────┴────────────────┴────────────────────────┴────────────┘

For every confirmed issue, a standalone pytest file is created automatically:

# test_exploit_vuln_tpa_001.py
def test_reproduce_exploit_vuln_tpa_001():
    """Auto-generated reproducible test verifying reachability of tool sink."""
    assert target_tool == "delete_file"
    assert captured_args == {"path": "/var/data/prod.db"}

🔒 2026 Core Enterprise Capabilities

1. $SECRET_n Privacy Tokenization Gateway

Separates reasoning from authority. Credentials never reach the LLM or logs:

from cybereiche_mcp_sec.sandbox import PrivacyGateway

gateway = PrivacyGateway()
gateway.register_secret("postgres://user:pass@db:5432/prod", name="DB_URL")

# Raw secrets are masked in prompt contexts and intercepted tool parameters
masked = gateway.mask_dict({"db": "postgres://user:pass@db:5432/prod"})
# Output: {"db": "$SECRET_DB_URL"}

2. Atomic Session Checkpointing & HITL Approval

Multi-hour runs are saved atomically with crash salvage:

from cybereiche_mcp_sec.fuzzer import FuzzingSessionManager, HITLApprovalGate

manager = FuzzingSessionManager(session_path=".cybereiche/session.json")
session = manager.load_or_create()

gate = HITLApprovalGate(on_approval_required=lambda tool, args: True)
assert gate.request_approval("delete_table", {"table": "users"}) is True

3. Closed-Loop Remediation Bridge

Verifies that code fixes neutralize the exploit without regressions:

from cybereiche_mcp_sec.remediation import FixVerificationBridge, RemediationStatus

bridge = FixVerificationBridge()
receipt = bridge.verify_remediation(
    proof=proof,
    pov_script_path=path_to_test_exploit,
    patch_applier=lambda: apply_git_patch("fix.patch"),
    regression_tester=lambda: run_full_test_suite(),
)

assert receipt.status == RemediationStatus.VERIFIED_FIX
assert receipt.is_deployable is True

🏢 GitHub Actions CI/CD Integration

Embed continuous MCP security verification into your pipeline:

name: CyberEiche MCP Security Gate
on: [push, pull_request]

jobs:
  mcp-security:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: astral-sh/setup-uv@v2
      - name: Run MCP Security Fuzzing
        run: uv run pytest --mcp-fuzz --mcp-sarif=results.sarif
      - name: Upload SARIF to GitHub Security Tab
        uses: github/codeql-action/upload-sarif@v3
        if: always()
        with:
          sarif_file: results.sarif

🌐 The CyberEiche Security Ecosystem

cybereiche-mcp-sec is part of the CyberEiche product suite:

  • cybereiche-vibesec: Autonomous AppSec, PostgreSQL RLS multi-tenant audit, and automated code fixes (vibesec fix).
  • cybereiche-mcp-sec: Deterministic MCP security verification, agent reachability fuzzing, and PoV generation.

🛡️ About CyberEiche

CyberEiche is a German cybersecurity firm based in Düsseldorf, specializing in enterprise defense, sovereign AI system security, and resilient cloud architectures.

Metadata

Release files for cybereiche-mcp-sec 0.1.2

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for cybereiche-mcp-sec 0.1.2
File Size Uploaded Provenance Actions
cybereiche_mcp_sec-0.1.2.tar.gz 20.2 kB Without provenance Details for cybereiche_mcp_sec-0.1.2.tar.gz

Built distribution (wheel)

Table of built distributions (wheels) for cybereiche-mcp-sec 0.1.2
File Interpreter ABI Platform Actions
cybereiche_mcp_sec-0.1.2-py3-none-any.whl Python 3 none any Details for cybereiche_mcp_sec-0.1.2-py3-none-any.whl

Total release size: 47.9 kB

Release files / cybereiche_mcp_sec-0.1.2.tar.gz

Download URL cybereiche_mcp_sec-0.1.2.tar.gz
Size 20.2 kB
Tags Source
SHA-256 checksum
How to use checksums
4975d562be983482cba196e251c07fdc5658d4d920d4d344c762e0853985de48
BLAKE2b-256 checksum
How to use checksums
f45471335c0121e89f2b9d0b65c74711c3ef2278abd0b1e680fe0120c264f632
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
No
Uploaded via twine/7.0.0 CPython/3.11.15

Release files / cybereiche_mcp_sec-0.1.2-py3-none-any.whl

Download URL cybereiche_mcp_sec-0.1.2-py3-none-any.whl
Size 27.7 kB
Tags Python 3
SHA-256 checksum
How to use checksums
d3ccfbfaa3337435321e2a3cb59c1e2f6d9a5f0fb03fd22dd40dfbb26641b4fd
BLAKE2b-256 checksum
How to use checksums
0b634ff5180da8fb85106d9c0c3f8d4672c3f3158c7e27ca8e2dada3f13fca1d
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
No
Uploaded via twine/7.0.0 CPython/3.11.15

Security for cybereiche-mcp-sec 0.1.2

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

No provenance found

None of the files in this release were uploaded with provenance metadata.

A note on package security

PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.

To learn more about protecting yourself from malicious actors, visit our package security documentation.

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page