Cyntrisec Cli MCP Server

Cyntrisec CLI

Cyntrisec Labs project. Cyntrisec's main product is EphemeralML — confidential AI inference with cryptographic receipts (AIR v1). This CLI is a standalone Labs tool. The PyPI package name cyntrisec and MCP server ID io.github.cyntrisec/cyntrisec are stable and will not change.

[!CAUTION] Beta Software Disclaimer: This tool is currently in BETA. It is provided "as is", without warranty of any kind. While Cyntrisec is a read-only analysis tool by default, the user assumes all responsibility for any actions taken based on its findings. Always review generated remediation plans and Terraform code before application.

AWS capability graph analysis and attack path discovery.

A read-only CLI tool that:

• Scans AWS infrastructure via AssumeRole
• Builds a capability graph (IAM, network, dependencies)
• Discovers attack paths from internet to sensitive targets
• Prioritizes fixes by ROI (security impact + cost savings)
• Identifies unused capabilities (blast radius reduction)
• Outputs deterministic JSON with proof chains

Demo

Watch how to discover attack paths and generate fixes using natural language with Claude MCP.

Architecture

+----------------------------------------------------------------------------------+
|                                   CYNTRISEC CLI                                   |
+----------------------------------------------------------------------------------+
| CLI Layer (Typer)                                                                 |
|   scan   analyze   cuts   waste   report   comply   can   diff   serve   ...      |
+-----------------------------+----------------------------------------------------+
| Core Engine                 | Storage (local)                                     |
|  - AWS collectors           |  ~/.cyntrisec/scans/<scan_id>/                      |
|  - Normalization/schema     |    snapshot.json, assets.json, relationships.json   |
|  - GraphBuilder -> AwsGraph |    findings.json, attack_paths.json                 |
|  - Path search -> paths     |  ~/.cyntrisec/scans/latest -> <scan_id>             |
|  - Min-cut + Cost (ROI)     |  (Windows fallback: latest is a file)               |
+-----------------------------+----------------------------------------------------+
| Outputs: JSON/agent, HTML report, remediation plan + Terraform hints              |
+----------------------------------------------------------------------------------+

Data Flow

CLI (scan) --AssumeRole--> AWS Session --Describe/Get/List--> AWS APIs (read-only)
     |
     v
Collectors -> normalize -> Assets + Relationships -> AwsGraph
                                                |
                                                v
                                   Attack path search (BFS/DFS)
                                                |
                                                v
                                   Min-cut (remediation cuts)
                                                |
                                                v
                                      Cost engine (ROI)

Local artifacts: ~/.cyntrisec/scans/<scan_id>/*.json

Installation

pip install cyntrisec

Windows PATH Fix

If you see "cyntrisec is not recognized", the Scripts folder isn't on PATH:

# Option 1: Run with python -m
python -m cyntrisec --help

# Option 2: Add to PATH for current session
$env:PATH += ";$env:APPDATA\Python\Python311\Scripts"

Quick Start

Prerequisite: Ensure you have AWS CLI installed and configured with credentials (e.g., aws configure) or environment variables set. terraform is required for the setup step.

# 1. Create the read-only IAM role in your account
cyntrisec setup iam 123456789012 --output role.tf

# 2. Apply the Terraform
cd your-infra && terraform apply

# 3. Run a scan
cyntrisec scan --role-arn arn:aws:iam::123456789012:role/CyntrisecReadOnly

# 4. View attack paths
cyntrisec analyze paths --min-risk 0.5

# 5. Find minimal fixes (prioritized by ROI)
cyntrisec cuts --format json

# 6. Generate HTML report
cyntrisec report --output report.html

Commands

Core Analysis

Setup & Validation

Remediation

Policy Testing

Agentic Interface

MCP Server Mode

Run Cyntrisec as an MCP server for AI agent integration:

# Install with MCP support (now included by default)
pip install cyntrisec

cyntrisec serve              # Start stdio server
cyntrisec serve --list-tools # List available tools

MCP Tools (15)

Claude Desktop

MacOS: ~/Library/Application Support/Claude/claude_desktop_config.json Windows: %APPDATA%\Claude\claude_desktop_config.json

{
  "mcpServers": {
    "cyntrisec": {
      "command": "python",
      "args": ["-m", "cyntrisec", "serve"]
    }
  }
}

Claude Code (CLI)

Run the following command to configure the server:

claude mcp add cyntrisec -- python -m cyntrisec serve

Google Gemini / Antigravity

Locate your agent configuration (e.g., ~/.gemini/antigravity/mcp_config.json) and add:

{
  "mcpServers": {
    "cyntrisec": {
      "command": "python",
      "args": ["-m", "cyntrisec", "serve"]
    }
  }
}

Trust & Safety

Read-Only Guarantees

This tool makes read-only API calls to your AWS account. The IAM role should have only Describe*, Get*, List* permissions.

No Data Exfiltration

All data stays on your local machine. Nothing is sent to external servers. Scan results are stored in ~/.cyntrisec/scans/.

No Auto-Remediation (Default Safe Mode)

By default, Cyntrisec is read-only and does not modify your AWS infrastructure.

• It analyzes your account using read-only APIs.
• It can generate remediation artifacts (e.g., Terraform modules) for you to review.
• It does not apply changes automatically.

Optional Remediation Execution (Explicit Opt-In)

Cyntrisec includes an explicitly gated path that can execute Terraform only if you intentionally enable it.

This mode is:

• Disabled by default
• Requires --enable-unsafe-write-mode
• Requires an additional explicit flag (e.g. --execute-terraform) to run Terraform
• Intended for controlled environments (sandbox / CI with approvals), not unattended production

If you do not pass these flags, Cyntrisec will never run terraform apply.

Write Operations

Cyntrisec makes no AWS write API calls during scanning and analysis.

The only supported "write" behavior is optional execution of Terraform locally on your machine, and only when explicitly enabled via unsafe flags.

Every AWS API call is logged in CloudTrail under session name cyntrisec-cli.

Trust & Permissions

Cyntrisec runs with a read-only IAM role. Generate the recommended policy with cyntrisec setup iam <ACCOUNT_ID> and keep permissions to Describe*, Get*, and List*. Live modes (waste --live, can --live) require extra IAM permissions; the generated policy and docs cover those additions.

Output Format

Primary output is JSON to stdout. When stdout is not a TTY, the CLI automatically switches to JSON:

cyntrisec analyze paths --format json | jq '.paths[] | select(.risk_score > 0.7)'

Agent-friendly output wraps results in a structured envelope:

cyntrisec analyze paths --format agent

{
  "schema_version": "1.0",
  "status": "success",
  "data": {...},
  "artifact_paths": {...},
  "suggested_actions": [...]
}

Exit Codes

Use in CI/CD:

cyntrisec scan --role-arn $ROLE_ARN || exit 1
cyntrisec diff || echo "Regressions detected"

Storage

Scan results are stored locally:

~/.cyntrisec/
|-- scans/
|   |-- 2026-01-17_123456_123456789012/
|   |   |-- snapshot.json
|   |   |-- assets.json
|   |   |-- relationships.json
|   |   |-- findings.json
|   |   `-- attack_paths.json
|   `-- latest -> 2026-01-17_...
`-- config.yaml

Versioning

This project follows Semantic Versioning. See CHANGELOG.md for release notes.

License

Apache-2.0

Links

• PyPI Package
• Website
• Twitter/X