Skip to main content
Glama
CyberSafe MCP

CyberSafe MCP

by kshitij9896

CyberSafe MCP (v0.3 — Hybrid AI/ML Edition)

An open-source, plain-language Model Context Protocol (MCP) server built in Python to help everyday users—especially people 40+—stay safe on phones, laptops, and AI tools.

Built with the official Python MCP SDK v2 (mcp>=2.0.0).

What's Inside v0.3

Hybrid AI/ML + Regional Rules Scam Detection

  • analyze_scam_message(message, channel, region) now runs a Hybrid Engine:

    1. Explainable Regional Packs (global, india, us_uk) in src/cybersafe_mcp/scam_packs/*.json for clear, zero-jargon red flags (Digital Arrest, UPI QR fraud, KYC traps, IRS/HMRC scams, AI voice-cloning).

    2. Local HuggingFace AI Model (optional ml extra) that catches subtle or paraphrased scams with no exact keywords—running 100% locally in memory with zero user data sent to third-party AI APIs.

Other Security Tools

  • check_url_safety(url) — Local phishing heuristics + optional live URLhaus API lookups (never visits the link).

  • check_email_breach(email) — Live breach check via the free, no-key XposedOrNot API.

  • get_latest_alerts(limit, consumer_only, keyword) — Live consumer-filtered alerts from the CISA KEV Catalog.

  • check_password_hygiene(password_sample) — Pure local offline password-strength evaluator.

  • get_daily_cyber_tip(category) — One 60-second daily tip (phone, laptop, ai, general).

Related MCP server: ScamShield

Quick Start

Lightweight Mode (Rules Only — Fastest Install)

uv sync --extra dev
uv run pytest
uv run mcp dev src/cybersafe_mcp/server.py

Full Hybrid AI/ML Mode (Rules + Local HuggingFace Model)

uv sync --extra dev --extra ml
uv run pytest
uv run mcp dev src/cybersafe_mcp/server.py

On the first call to analyze_scam_message, the ~17.6 MB BERT-Tiny model downloads once into your local HuggingFace cache and runs completely offline after that.

To swap models without changing code:

# macOS / Linux
CYBERSAFE_ML_MODEL=Horizon-Labs/phishing-scam-detector-base uv run mcp dev src/cybersafe_mcp/server.py

# Windows PowerShell
$env:CYBERSAFE_ML_MODEL="Horizon-Labs/phishing-scam-detector-base"
uv run mcp dev src/cybersafe_mcp/server.py

Remote Deployment (Streamable HTTP)

uv run cybersafe-mcp --transport streamable-http --host 0.0.0.0 --port 8000

Health endpoint: GET /health (also reports whether the AI/ML engine is enabled and which model is configured).

Available Tools

6 tools
analyze_scam_messageAInspect

Check a suspicious SMS, WhatsApp, email, or call script for scam red flags.

Combines regional rule packs with an optional local HuggingFace AI scam model.

Args: message: The exact text of the suspicious message or call script. channel: Where it arrived (e.g. 'SMS', 'WhatsApp', 'Email', 'Phone Call'). region: Regional scam pack to apply ('all', 'global', 'india', or 'us_uk').

Inputs: message (required), region, channel. Returns: result

check_email_breachAInspect

Check if an email address appeared in known public data breaches via XposedOrNot.

Args: email: The email address to check (never logged or stored).

Inputs: email (required). Returns: result

check_password_hygieneAInspect

Check password strength locally in memory without logging or sending it anywhere.

Tip: Test a similar pattern rather than pasting your real banking password.

Args: password_sample: The password or sample pattern to evaluate.

Inputs: password_sample (required). Returns: result

check_url_safetyAInspect

Check whether a suspicious web link looks like a phishing or malware URL.

Combines local heuristic checks with live abuse.ch URLhaus lookups (when URLHAUS_AUTH_KEY is set). Never visits or opens the submitted link.

Args: url: The web link to inspect (for example, 'https://example.com/login').

Inputs: url (required). Returns: result

get_daily_cyber_tipAInspect

Get one short, jargon-free daily cybersecurity tip.

Args: category: One of 'phone', 'laptop', 'ai', or 'general'.

Inputs: category. Returns: result

get_latest_alertsAInspect

Get recent actively exploited security alerts from the free CISA KEV catalog in plain English.

Args: limit: Number of alerts to return (1 to 15, default 5). consumer_only: If True, focuses on everyday devices & browsers (Apple, Google, Microsoft, etc.). keyword: Optional product or vendor filter (e.g. 'Chrome', 'iPhone', 'Windows').

Inputs: limit, keyword, consumer_only. Returns: result

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    TDQS

    A4/5.0

    Scored across 6 tools

    Disambiguation5/5

    Each tool targets a clearly distinct artifact: a URL, an email address, a message transcript, a threat feed, a password, and a daily tip. There is no overlap in inputs or purpose, so an agent can select the right tool from the input type alone.

    Naming Consistency5/5

    All six tools follow a consistent snake_case verb_noun pattern (check_url_safety, get_latest_alerts, analyze_scam_message, etc.). The check_/analyze_/get_ prefixes are all conventional verb forms and remain predictable across the set.

    Tool Count5/5

    Six tools is well-scoped for a consumer cybersecurity helper; each tool covers a distinct check with no redundancy. Nothing feels padded or missing at the structural level.

    Completeness4/5

    The surface covers the core consumer safety checks: link, email breach, message scam, vulnerability alerts, password, and tips. Minor gaps exist for adjacent checks such as IP/domain reputation or phone-number breach lookup, but agents can work around them.

    Maintenance

    ActivityMaintained
    ResponsivenessNo recent community issues

    Related MCP Connectors

    Related MCP Servers

    • A
      license
      Not graded
      quality
      B
      maintenance
      Enables AI assistants to autonomously detect scams by checking URLs, messages, ABNs, and phone numbers using Australian-specific threat intelligence and privacy-preserving techniques.
      MIT
    • A
      license
      Not graded
      quality
      B
      maintenance
      Enables Alexa+ and other MCP-capable assistants to voice-check suspicious messages, links, and phone numbers, with guided call interviews, official-source verification, and plain-language scam briefings.
      MIT
    • A
      license
      Not graded
      quality
      C
      maintenance
      Enables users to check suspicious text, emails, marketplace listings, job offers, investment pitches, and links through MCP tools, returning an assessment with supporting checks and next-step guidance.
      Apache 2.0
    • F
      license
      Not graded
      quality
      C
      maintenance
      Enables checking suspicious emails, SMS, WhatsApp messages, links or screenshot descriptions for scam signals targeting people in the Netherlands and Belgium, returning a verdict, the red flags found, the impersonated brand and next-step advice. Exposes a single check_message tool over a hosted Streamable HTTP endpoint.
      -