⚡🛡️ EchoPulse & ConsentGate MCP — Multimodal Acoustic Home Triage & Information-Flow Firewall for Alexa+

"Hear the mechanical fault. See the X-Ray blueprint. Stage the exact $28 Amazon Prime OEM part. Block lethal hazards and Confused Deputy prompt injections at the MCP transport boundary."

🌐 Try It Immediately (Zero Setup Required)

• 🖥️ Live Interactive Echo Show 15 Wall Hub (GitHub Pages): https://jasurcyb.github.io/echopulse-consentgate-mcp/

  • Includes Studio Neural AI Voiceover (en-US-AvaNeural & en-US-GuyNeural), 8 Physics-Synthesized Appliance & Hazard Presets, Live Microphone FFT Capture (🎙️ Live Mic), Interactive SVG X-Ray Appliance Blueprints, Confused Deputy Attack Arena, and Live MCP 2025-11-25 JSON-RPC Inspector.

• ⚡ Local Streamable HTTP MCP Server + Automated Verification:

  git clone https://github.com/Jasurcyb/echopulse-consentgate-mcp.git
  cd echopulse-consentgate-mcp
  npm test            # Runs 9 automated verification suites in < 550ms
  npm run benchmark   # Executes all 220 quantitative evaluations (112 PCM waveforms + 108 IFC scenarios)
  npm start           # Starts MCP 2025-11-25 Streamable HTTP server + Web UI on http://localhost:8080

Related MCP server: open-edison

🏛️ Hackathon Track & Mini-Challenge Verification Matrix

For judges reviewing across the primary track and mini-challenges, every requirement maps directly to self-contained, verifiable modules in this repository:

🎯 Why EchoPulse & ConsentGate Exists

Every year, 68 million U.S. households hear an unfamiliar mechanical sound from a washing machine, refrigerator, furnace, or water main—and face a costly dilemma:

1. Pay $185–$280 just for an emergency technician visit to diagnose a $22–$39 OEM part they could replace in 20 minutes.

2. Ask a baseline text LLM, which fails in two critical ways:

   • Deaf to Mechanical Physics: Text LLMs cannot compute a 2048-point Fast Fourier Transform, measure Total Harmonic Distortion ($\text{THD}%$), distinguish a $2,424\text{ Hz}$ outer-race bearing screech from a $180\text{ Hz}$ rotor hum, or detect a lethal 240V Contactor Plasma Arc.

   • Vulnerable to Confused Deputy Hijacking: When an autonomous Alexa+ agent scrapes third-party repair bulletins, calendar invites, emails, or courier notes over MCP, hidden Indirect Prompt Injections can hijack high-consequence tools—ordering counterfeit $349 parts (place_amazon_order), unlocking smart deadbolts (unlock_front_door), or disarming Ring alarms (disarm_ring_alarm).

EchoPulse & ConsentGate MCP solves both problems deterministically:

1. 🎧 Real-Time 2048-Point Radix-2 Cooley-Tukey FFT (engine/acoustic-dsp.js): Extracts exact harmonic signatures from live microphone audio or synthesized mechanical waveforms, identifies the failing component, and stages the exact Amazon Prime OEM replacement part (#WPW10348269 — $28.40, saving $186.60).

2. 📐 Interactive X-Ray Appliance Blueprints: Renders animated SVG cross-sections on the Echo Show 15 Wall Hub pinpointing the faulty sub-assembly in real time.

3. 🛡️ Deterministic Hazard Lockout & Information-Flow Control Firewall (engine/lattice.js):

   • Automatically locks out DIY instructions and trips smart breakers/valves when 240V Electrical Arcing or UL-2034 Carbon Monoxide Temporal-4 ($3,104\text{ Hz}$) hazards are detected.

   • Enforces formal capability triples $(\tau, \kappa, \sigma)$ across all 16 Alexa+ MCP tools, intercepting Confused Deputy attacks at the MCP transport boundary with a visual Provenance Diff and 2-Word Phonetic Voice Step-Up Challenge ("Echo Falcon").

🏗️ System Architecture

flowchart TB
    subgraph EchoShow["🖥️ Amazon Echo Show 15 Wall Hub (Voice + Mic + 60 FPS Display)"]
        Mic["🎙️ 16.384 kHz Live Mic / Web Audio Physics Synth"]
        Voice["🗣️ Verified User Voice Utterance (Intent Anchor)"]
        Screen["📐 Live 60 FPS FFT Spectrogram + X-Ray Blueprint + Voice Step-Up UI"]
    end

    subgraph MCPServer["⚡ EchoPulse & ConsentGate MCP Server (Spec 2025-11-25 Streamable HTTP)"]
        FFT["1. Radix-2 Cooley-Tukey FFT & Blackman-Harris Spectral Classifier"]
        Hazard["2. Life-Safety Hazard Lockout (240V Plasma Arc & UL-2034 CO T4)"]
        Lattice["3. ConsentGate IFC Taint Lattice (τ, κ, σ) & Provenance Diff"]
    end

    subgraph Ecosystem["🏠 Amazon Home, Commerce & Cloud Ecosystem"]
        Prime["🛒 Amazon Prime Verified OEM Part Staging (#WPW10348269)"]
        SmartHome["🔌 Ring Alarm, Smart Locks, Smart Water Valve & 240V Breaker"]
        AWS["☁️ Amazon Bedrock Converse API + DynamoDB Audit Ledger"]
    end

    Mic --> FFT
    Voice --> Lattice
    FFT --> Hazard
    Hazard -->|"Safe DIY Fault"| Prime
    Hazard -->|"Lethal 240V / CO T4"| SmartHome
    FFT --> Screen
    Lattice -->|"BLOCKED / STEP_UP_REQUIRED"| Screen
    Lattice -->|"ALLOWED (Clean or Voice Token Verified)"| SmartHome
    Lattice --> AWS

🔬 Mathematical & Algorithmic Formulation

1. 4-Term Blackman-Harris Windowed Radix-2 FFT (engine/acoustic-dsp.js)

To eliminate spectral leakage between adjacent $8\text{ Hz}$ frequency bins ($f_s = 16,384\text{ Hz}, N = 2048$), each time-domain frame $x[n]$ is weighted by the 4-term Blackman-Harris window ($-92\text{ dB}$ sidelobe floor) prior to in-place Cooley-Tukey bit-reversal decimation:

$$w[n] = 0.35875 - 0.48829\cos\left(\frac{2\pi n}{N-1}\right) + 0.14128\cos\left(\frac{4\pi n}{N-1}\right) - 0.01168\cos\left(\frac{6\pi n}{N-1}\right)$$

From the Welch-averaged power spectrum $P[k] = |X[k]|^2$, EchoPulse extracts:

• Total Harmonic Distortion ($\text{THD}%$): $$\text{THD}(%) = 100 \times \frac{\sqrt{\sum_{m=2}^{6} P[m \cdot k_0]}}{P[k_0]}$$

• Time-Domain Crest Factor ($\text{dB}$) & Pulse Duty Ratio: Distinguishes continuous mechanical hums from transient copper-pipe water-hammer shockwaves ($\text{Crest} > 11.5\text{ dB}$) and UL-2034 Temporal-4 Carbon Monoxide alarms ($3,104\text{ Hz}$ carrier with 4-chirp cadence).

• High-Frequency Plasma/Arc Energy Ratio ($R_{\text{HF}}$): Detects broadband $>3.5\text{ kHz}$ sputter characteristic of 240V contactor arcing ($R_{\text{HF}} > 8.5%$) combined with $120/180/300\text{ Hz}$ line-frequency intermodulation.

2. ConsentGate Information-Flow Control (IFC) Lattice (engine/lattice.js)

Every MCP tool $T_i$ is assigned a static capability triple $(\tau_i, \kappa_i, \sigma_i)$ over a bounded lattice $(\mathcal{L}, \sqsubseteq)$:

• Trust Level $\tau \in {0:\text{UNTRUSTED_EXTERNAL},; 1:\text{SEMI_TRUSTED_SKILL},; 2:\text{VERIFIED_USER_VOICE}}$

• Consequence Class $\kappa \in {0:\text{READ_ONLY},; 1:\text{REVERSIBLE_WRITE},; 2:\text{CRITICAL_EXTERNAL},; 3:\text{PHYSICAL_FINANCIAL}}$

• Scope Domain $\sigma \in {\text{INFO}, \text{DIAGNOSTICS}, \text{COMMERCE}, \text{FINANCE}, \text{PHYSICAL_ACCESS}, \text{LIFE_SAFETY}, \text{COMMS}, \text{AUTOMATION}}$

As an Alexa+ session executes a sequence of tool calls $T_1, T_2, \dots, T_t$, the session context taint evolves monotonically:

$$\tau_{\text{ctx}}^{(t)} = \max\left(\tau_{\text{ctx}}^{(t-1)},; \tau_{\text{source}}(T_t)\right)$$

Whenever a tool with $\kappa(T_t) \ge 2$ is invoked while $\tau_{\text{ctx}}^{(t)}$ is tainted by untrusted data (fetch_repair_bulletin, read_calendar_invite, read_delivery_instructions, read_email) and lacks an explicit user Intent Anchor ($T_t \notin \mathcal{A}(u_0)$), execution is deterministically halted at the MCP boundary until the user speaks a cryptographic HMAC-SHA256 2-Word Voice Step-Up Token.

📊 Dual Reproducible Benchmark Results (220 Total Evaluations)

Run npm run benchmark to execute both quantitative suites from scratch in under 250ms and regenerate data/results.json:

Suite A: Radix-2 FFT Acoustic Fault Classification Across AWGN Noise Tiers (112 PCM Waveforms)

Evaluates 16,384-sample floating-point PCM waveforms synthesized across 7 physical fault & baseline classes (washer_drain_pump_cavitation, fridge_evap_fan_frost_rub, hvac_capacitor_60hz_hum, furnace_inducer_bearing_wear, plumbing_water_hammer_transient, electrical_panel_240v_arc_hazard, carbon_monoxide_ul2034_temporal4) at 4 calibrated Additive White Gaussian Noise (AWGN) SNR levels (+20 dB, +12 dB, +6 dB, 0 dB × 4 seeds):

Suite B: Cross-Tool Indirect Prompt Injection Benchmark (108 Multi-Turn Scenarios)

Evaluates 72 adversarial Confused Deputy attacks across 6 families (calendar_trojan, delivery_note_hijack, commerce_skimming, pii_exfiltration, routine_persistence, acoustic_tv_spoof) + 36 legitimate multi-step household controls:

🔌 Live MCP 2025-11-25 Streamable HTTP API Verification

Once npm start is running on http://localhost:8080, you can verify the Model Context Protocol (2025-11-25) endpoint directly via curl:

1. Initialize an MCP 2025-11-25 Session

curl -i -X POST http://localhost:8080/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":1,"method":"initialize","params":{"protocolVersion":"2025-11-25","clientInfo":{"name":"alexa-plus-verifier","version":"1.0.0"}}}'

2. Run an Acoustic FFT Appliance Diagnostic (echopulse_scan_appliance)

curl -s -X POST http://localhost:8080/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":2,"method":"tools/call","params":{"name":"echopulse_scan_appliance","arguments":{"presetId":"washer_drain_pump_cavitation","snrDb":18},"_meta":{"utterance":"Alexa, listen to my noisy washing machine and diagnose the part"}}}'

3. Verify Confused Deputy Interception (CONSENTGATE_INTERCEPT)

curl -s -X POST http://localhost:8080/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":3,"method":"tools/call","params":{"name":"read_delivery_instructions","arguments":{"packageId":"TBA-992"},"_meta":{"sessionId":"demo-hijack","utterance":"Alexa, check my delivery instructions"}}}'

curl -s -X POST http://localhost:8080/mcp \
  -H "Content-Type: application/json" \
  -d '{"jsonrpc":"2.0","id":4,"method":"tools/call","params":{"name":"unlock_front_door","arguments":{"lockId":"front-deadbolt"},"_meta":{"sessionId":"demo-hijack"}}}'

🚀 Quickstart (Zero Dependencies — Node.js Built-ins Only)

In strict compliance with supply-chain security standards, this repository uses 0 external npm packages (package.json has zero dependencies or devDependencies—100% immune to third-party npm supply-chain compromise).

# 1. Run the automated test suite (9 test suites covering FFT DSP, IFC Lattice, 220 benchmarks, and MCP 2025-11-25)
npm test

# 2. Run the dual quantitative benchmark (112 PCM waveforms + 108 security scenarios)
npm run benchmark

# 3. Start the Streamable HTTP MCP 2025-11-25 Server + Echo Show 15 Wall Hub UI
npm start
# Open http://localhost:8080 in your browser (or open index.html directly!)

📂 Repository Structure

echopulse-consentgate-mcp/
├── engine/
│   ├── acoustic-dsp.js        # 2048-pt Radix-2 Cooley-Tukey FFT, Blackman-Harris window & PCM fault synth
│   └── lattice.js             # Deterministic Information-Flow Control (IFC) Lattice (16 Alexa+ MCP tools)
├── data/
│   ├── corpus.js              # 108 multi-turn scenarios (72 attacks across 6 families + 36 controls)
│   └── results.json           # Generated dual benchmark output (112 PCM waveforms + 108 IFC traces)
├── mcp/
│   ├── server-core.js         # Universal MCP Spec 2025-11-25 Engine (Streamable HTTP & In-Browser Loopback)
│   └── aws-adapter.js         # Amazon Bedrock Converse & DynamoDB audit adapter (zero-key offline fallback)
├── audio/                     # 11 studio-grade neural AI voiceover assets (en-US-AvaNeural & GuyNeural)
├── scripts/
│   └── run-benchmark.js       # Dual benchmark runner (Acoustic FFT SNR + Cross-Tool IFC + Wilson CIs)
├── tests/
│   └── consentgate.test.js    # Node.js native automated verification suite (9 test suites via node --test)
├── index.html                 # Echo Show 15 Acoustic & X-Ray Wall Hub + Confused Deputy Arena
├── css/styles.css             # Echo Show 15 hardware bezel, 60 FPS canvas & X-Ray SVG animations
├── js/app.js                  # 60 FPS spectrogram renderer, Web Audio synth, live mic & MCP inspector
├── server.js                  # Streamable HTTP MCP Server (POST /mcp, GET /mcp SSE, DELETE /mcp)
├── SKILL.md                   # Open Agent Skill manifest for Alexa+
├── FRICTION-LOG.md            # Engineering feedback & RFC for Amazon Alexa+ / MCP teams
├── vercel.json                # Zero-config deployment manifest
└── LICENSE                    # MIT License