KrishiArtha-MCP (MandiPulse) 🌾

Autonomous Covariate Mandi Arbitrage, Weather-Spoilage Physics & Statutory Distress-Credit Engine

A Multi-Agent Neuro-Symbolic Platform for Indian Agriculture

🏆 FORGE'26 Hackathon Submission Overview

• Team: INFINITE LOOP (IIIT Guwahati)

• Members: Abhilok Reddy Bellamkonda (Leader) & Nimmakayala Naga Venkata Shanmukha Sai Chandra

• Chosen Track: Climate, Resource Infrastructure & Global Supply Resilience (with cross-cutting Digital Public Infrastructure & Governance impact)

🎯 Fast Links for Judges & Evaluators

Related MCP server: Agmarknet MCP Server

📌 The Problem

Smallholder farmers and Farmer Producer Organizations (FPOs) across India lose 25% to 40% of net crop value during unseasonal weather shifts and harvest gluts.

Existing government portals (Agmarknet 2.0, eNAM) fail because:

1. Backward-Looking: They display historical prices but cannot predict price crashes under approaching weather shocks.

2. Ignorant of Spoilage: They assume 0% crop loss, ignoring that hauling perishable commodities 200+ km under 32°C heat and 85% humidity in open tarp trucks causes rapid biological rot ($Q_{10}$ microbial decay).

3. Zero Safety Net: When local prices crash below production costs, portals offer zero actionable financial recourse.

💡 The Solution: Neuro-Symbolic Multi-Agent Architecture

KrishiArtha-MCP combines foundation time-series AI with thermodynamic transit decay physics and statutory warehouse pledge financing into an autonomous 4-node LangGraph pipeline:

flowchart TB
    subgraph Client["Frontend Edge (Streamlit / Bhashini)"]
        Voice["Bhashini Multilingual Speech / Audio Triggers (Marathi, Hindi, Telugu, English)"]
        UI["Mission Control Cockpit & 3-Column Arbitrage Matrix"]
        ShockSlider["What-If Unseasonal Rainfall Shock Slider (+0mm to +100mm)"]
    end

    subgraph FastMCP["FastMCP Ingestion & Telemetry"]
        MandiAPI["mandi-api.onrender.com (Live APMC Prices)"]
        WeatherAPI["Open-Meteo API (7-Day Rain/Temp Covariates)"]
        OSRMAPI["OSRM Routing (Highway Distance & Geometry)"]
    end

    subgraph CoreEngine["Quantitative Foundation Models & Physics"]
        Chronos["amazon/chronos-2 (P10, P50, P90 Quantiles)"]
        Spoilage["Arrhenius Q10 Decay + Round-Trip Diesel Math"]
    end

    subgraph Governance["LangGraph v1.2.18 Stateful Orchestration"]
        Node1["Node 1: quant_arbitrage"]
        Node2["Node 2: policy_graphrag (LightRAG Hybrid Mode)"]
        Gate["Node 3: human_approval_gate (interrupt() Consent Gate)"]
        Node4["Node 4: execute_enwr_browser (Playwright WDRA Agent)"]
    end

    Voice --> UI
    UI --> Node1
    Node1 <--> FastMCP
    FastMCP --> MandiAPI & WeatherAPI & OSRMAPI
    Node1 <--> CoreEngine
    CoreEngine --> Chronos & Spoilage
    Node1 --> Node2
    Node2 --> Gate
    Gate -- "Command(resume=True)" --> Node4
    Node4 --> UI

🔬 Mathematical & Physical Defensibility

1. Arrhenius $Q_{10}$ Biological Transit Decay

$$k(T_{\text{amb}}, RH) = k_{\text{base}} \cdot \left(Q_{10}\right)^{\frac{T_{\text{amb}} - 20.0}{10.0}} \cdot \left(1 + \alpha_{\text{hum}} \cdot \max\left(0, \frac{RH - 70}{100}\right)\right)$$

• $Q_{10} = 2.2$, $k_{\text{base}} = 0.0065$ (Tomato), $\alpha_{\text{hum}} = 1.5$.

• Total transit time: $t_{\text{transit}} = t_{\text{duration}} + 2.5\text{h}$ mandatory APMC unloading queue buffer.

• Spoilage loss fraction: $L = 1 - \exp\left(-k \cdot t_{\text{transit}}\right)$.

2. Freight & Net Realizable Price ($\text{NRP}$)

$$C_{\text{transport}} = 2 \cdot d_{\text{km}} \cdot ₹22/\text{km} + Q \cdot ₹35/\text{qtl}$$ $$\text{NRP}m = \left[ P{50} \cdot (1 - L) \cdot (1 - \gamma_{\text{APMC}}) \right] - \frac{C_{\text{transport}}}{Q}$$

📊 The 3-Column Financial Arbitrage Matrix (Nashik 50 qtl Lot)

🛡️ Security Audit (AgentShield)

Audited via npx ecc-agentshield scan:

• Grade: A (100/100)

• Findings: 0 Critical, 0 High, 0 Medium, 0 Low, 0 Info

• Secrets: 100/100 (Zero exposed API keys or tokens)

• Permissions: 100/100 (Safe sandbox compliance)

⚡ Quick Start

1. Clone & Install Dependencies

git clone https://github.com/0xabhilok/krishiartha-mcp.git
cd krishiartha-mcp
pip install -r requirements.txt

2. Run the Full Test Suite

# Run all 70 unit tests in offline stage safe mode
pytest

3. Launch the Mission Control Cockpit

streamlit run frontend/app.py

Open http://localhost:8501 in your browser, or visit the live deployment at https://krishiartha-mcp.streamlit.app.

📂 Project Organization

• contracts/schemas.py: Frozen Pydantic v2 schemas.

• mcp_server/: FastMCP server, Chronos-2 quantiles, $Q_{10}$ transit spoilage math.

• policy_rag/: LightRAG hybrid knowledge graph & NABARD/WDRA circular citations.

• orchestrator/: 4-node LangGraph StateGraph, interrupt() consent gate, Playwright WDRA agent.

• frontend/: Streamlit dashboard, Bhashini multilingual speech triggers, 3-column matrix.

• data/offline_snapshots/: Air-gapped deterministic stage snapshot for zero-fail live demos.

• docs/: 8-Slide World-Class Pitch Deck (Native Morph), 5-Minute Video Walkthrough Script, and Submission Copy.