CyberEiche MCP-Sec (cybereiche-mcp-sec)
Enterprise-grade reachability security testing & native
pytestplugin for Anthropic Model Context Protocol (MCP) servers and AI agent clients — engineered by CyberEiche.
Recent empirical security research reveals that 87.5% of verified MCP CVEs are High or Critical severity, with 66.7% leading to direct code execution or path traversal via Tool Poisoning Attacks (TPAs).
Traditional SAST and SCA tools generate 80–90% false-positive noise by reporting theoretical vulnerabilities. cybereiche-mcp-sec applies DARPA AIxCC principles—static reachability graph analysis, directed multi-turn conversational fuzzing (Crescendo / TAP), and ephemeral in-memory sandboxing—to deliver a 0% false-positive guarantee.
An alert is raised if and only if an executable, reproducible Proof-of-Vulnerability script (test_exploit.py) is synthesized.
🏗️ Architecture
┌────────────────────────────────────────────────────────────────────────────────────────┐
│ CYBEREICHE MCP-SEC SECURITY TESTING PIPELINE │
├────────────────────────────────────────────────────────────────────────────────────────┤
│ │
│ [Module 1: Static Tool Reachability Graph] │
│ Parses MCP server definitions (`tools/list`) and identifies dangerous tool sinks │
│ (FILESYSTEM, COMMAND_EXEC, DATABASE, NETWORK) and unvalidated parameters. │
│ │ │
│ ▼ │
│ [Module 2: Directed Multi-Turn Fuzzer (TAP & Crescendo)] │
│ Executes adaptive conversational attacks against agents to steer them toward sinks. │
│ │ │
│ ▼ │
│ [Module 3: Ephemeral Micro-Sandbox with Canary Oracles] │
│ Intercepts JSON-RPC tool calls safely in memory and plants cryptographic canaries. │
│ │ │
│ ▼ │
│ [Module 4: Deterministic PoV Synthesizer & Pytest Plugin] │
│ Generates `test_exploit_<id>.py` reproduction scripts and SARIF 2.1.0 reports. │
│ │
└────────────────────────────────────────────────────────────────────────────────────────┘
⚡ Quickstart
1. Installation
# Via uv (recommended)
uv add --dev cybereiche-mcp-sec
# Via pip
pip install cybereiche-mcp-sec
2. Run with Pytest
Execute security fuzzing against any configured MCP server or target agent:
# Run security suite and output SARIF report for GitHub Actions
pytest --mcp-fuzz --mcp-sarif=mcp-security-results.sarif
3. Beautiful Terminal Scorecard
CyberEiche MCP-Sec outputs clean, actionable results with zero noise:
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│
└──────────────────┴──────────┴────────────────┴────────────────────────┴────────────┘
🛡️ The Zero-Noise Guarantee
cybereiche-mcp-sec enforces the DARPA AIxCC Zero-Noise Principle:
- Never alert on speculation: Abstract code patterns or theoretical prompt injections without tool execution are never flagged as vulnerabilities.
- Deterministic PoV Output: For every verified 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 Architectural Capabilities
1. $SECRET_n Privacy Tokenization Gateway
Separates reasoning from authority and secrets. Before any tool arguments or conversational context are routed, PrivacyGateway tokenizes sensitive credentials, GitHub tokens, and connection strings into $SECRET_1, $SECRET_2:
from cybereiche_mcp_sec.sandbox import PrivacyGateway
gateway = PrivacyGateway()
gateway.register_secret("postgres://user:pass@db:5432/prod", name="DB_URL")
# Raw secrets are masked in prompt contexts and intercepted tool parameters
masked = gateway.mask_dict({"db": "postgres://user:pass@db:5432/prod"})
# Output: {"db": "$SECRET_DB_URL"}
2. Resumable Sessions & Human-in-the-Loop (HITL) Approval Gates
Multi-day engagements and long-running security evaluations support checkpointing and crash salvage:
- Session Persistence: State is auto-saved to
.cybereiche/fuzz_session.jsonto avoid re-running already completed scenarios. - HITL Approval Gate: Prevents destructive tools (
delete_file,drop_database) from executing unchecked unless approved by operator policy (ApprovalPolicy.CALLBACKorApprovalPolicy.AUTO_APPROVE).
3. Closed-Loop Remediation & Fix Verification Bridge
Proves an exploit has been eliminated without introducing 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_code_patch(), # e.g. vibesec fix
regression_tester=lambda: run_full_test_suite(),
)
assert receipt.status == RemediationStatus.VERIFIED_FIX
assert receipt.is_deployable is True
🧩 Programmatic Python API
from cybereiche_mcp_sec.parser import MCPParser
from cybereiche_mcp_sec.sandbox import MockMCPServer, CanaryOracle
from cybereiche_mcp_sec.fuzzer import FuzzingEngine
# 1. Parse MCP server schema
parser = MCPParser()
catalog = parser.parse_config_json(server_config_json)
# 2. Spin up safe in-memory mock server with canary oracle
oracle = CanaryOracle()
token = oracle.plant_virtual_file("/etc/shadow")
server = MockMCPServer(tools=catalog.servers["filesystem"].tools, canary_oracle=oracle)
# 3. Execute multi-turn adversarial probe
fuzzer = FuzzingEngine()
result = fuzzer.execute_scenario(agent_under_test, server, scenario)
if result.vulnerability_confirmed:
print(f"Verified Exploit Found: {result.proof.title}")
🤝 CI/CD & SARIF Integration
Add CyberEiche MCP-Sec directly to your GitHub Actions workflow:
name: CyberEiche MCP Security Audit
on: [push, pull_request]
jobs:
security:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v4
- uses: astral-sh/setup-uv@v2
- run: uv run pytest --mcp-fuzz --mcp-sarif=results.sarif
- uses: github/codeql-action/upload-sarif@v3
if: always()
with:
sarif_file: results.sarif
🛡️ About CyberEiche
CyberEiche is a German cybersecurity firm specializing in enterprise defense, resilient infrastructure, and AI system security.
- Website: cybereiche.de
- Contact: contact@cybereiche.de
- License: MIT License
Metadata
Release files for cybereiche-mcp-sec 0.1.1
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| cybereiche_mcp_sec-0.1.1.tar.gz | 19.8 kB | Without provenance | Details for cybereiche_mcp_sec-0.1.1.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| cybereiche_mcp_sec-0.1.1-py3-none-any.whl | Python 3 | none | any | Details for cybereiche_mcp_sec-0.1.1-py3-none-any.whl |
Total release size: 47.3 kB
Release files / cybereiche_mcp_sec-0.1.1.tar.gz
| Download URL | cybereiche_mcp_sec-0.1.1.tar.gz |
|---|---|
| Size | 19.8 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
96f65da075380992abdefa71a2b2b2cc88b256783aba5137dca2e937382da6bd
|
|
BLAKE2b-256 checksum How to use checksums |
b4342b47709395029f14b69af4a1104fb69d6ba67496d5a3ddde95f3b6c10a8a
|
| 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-0.1.1-py3-none-any.whl
| Download URL | cybereiche_mcp_sec-0.1.1-py3-none-any.whl |
|---|---|
| Size | 27.4 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
4d7f8a92675b458633b18d36da46e01b69fbcc7498defeb9412bc278050b8e38
|
|
BLAKE2b-256 checksum How to use checksums |
e34f87510303acdca773bf14a92fe94779dd5ea83efe26da39fb70af2d92cd88
|
| 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 0.1.1
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.