# DataCompute Agent — AI-Powered Compute-Over-Data on Filecoin

> Autonomous MCP server that fetches datasets from IPFS/Filecoin, performs computation (anomaly detection, statistical analysis, data quality scoring), and stores results back via Multi-Chain Storage (MCS) simulation.

Built for **Data DAO Hackathon** (Filecoin/DoraHacks) — **Tracks 2 & 3** ($35K combined).

## 🎯 Problem

Data stored on Filecoin/IPFS is growing rapidly, but performing computation on that data requires manual retrieval, local processing, and re-upload — a slow, fragmented workflow. There's no autonomous agent layer that can fetch datasets, perform analysis, and store verified results back on-chain.

## ✅ Solution

**DataCompute Agent** is an MCP (Model Context Protocol) server with 6 callable tools that provides autonomous compute-over-data on Filecoin/IPFS datasets:

1. **Fetch** datasets from IPFS by CID via public gateways
2. **Compute** statistics (mean, median, std dev, quartiles)
3. **Detect** anomalies using Z-score and IQR methods
4. **Score** data quality (completeness, uniqueness, consistency)
5. **Store** results to Filecoin via MCS (Multi-Chain Storage) simulation
6. **Pipeline** — run the complete fetch → compute → detect → quality → store cycle

## 🏆 Unique Angle

> Unlike existing Filecoin MCP servers that only handle storage (foc-storage-mcp, storacha/mcp), DataCompute Agent adds a **computation layer** — it doesn't just store/retrieve data, it *analyzes* data in-transit and stores verified results back on-chain. First MCP server to combine compute-over-data with multi-chain storage on Filecoin.

## 📐 Architecture

```
IPFS/Filecoin (Storage)              DataCompute Agent (MCP)
+-----------------------+           +--------------------------+
|  Dataset (CID)         |          |  1. fetch_dataset()      |
|  CSV, JSON, JSONL     |--------->|  2. compute_statistics()  |
|                        |          |  3. detect_anomalies()   |
|  Results (via MCS)    |<---------|  4. data_quality_score()  |
|  CID pinning          |          |  5. store_results()       |
+-----------------------+          |  6. full_pipeline()       |
                                    +--------------------------+
                                             |
                                             v
                                    +--------------------------+
                                    |  FastAPI Web Dashboard    |
                                    |  - Submit CID for analysis |
                                    |  - View results & reports  |
                                    |  - MCP tool explorer       |
                                    +--------------------------+
```

## 🛠️ MCP Tools (6 callable)

| # | Tool | Description | Endpoint |
|---|------|-------------|----------|
| 1 | `fetch_dataset` | Retrieve dataset from IPFS by CID | `GET /api/tools/fetch?cid=<CID>` |
| 2 | `compute_statistics` | Descriptive statistics on columns | `GET /api/tools/statistics?cid=<CID>` |
| 3 | `detect_anomalies` | Z-score / IQR anomaly detection | `GET /api/tools/anomalies?cid=<CID>&method=zscore` |
| 4 | `data_quality_score` | Completeness, uniqueness, consistency | `GET /api/tools/quality?cid=<CID>` |
| 5 | `store_results` | Store results via MCS simulation | `POST /api/tools/store` |
| 6 | `full_pipeline` | Complete fetch→compute→detect→store | `POST /api/tools/pipeline` |

## 🚀 Setup

```bash
git clone https://github.com/0xConsole/datacompute-agent.git
cd datacompute-agent
pip install -r requirements.txt
uvicorn main:app --reload --port 8000
```

Open `http://localhost:8000` for the dashboard.

## 🌐 Live Demo

**URL:** https://datacompute-agent.vercel.app

## 📊 Tech Stack

| Component | Technology |
|-----------|-----------|
| Backend | Python + FastAPI |
| MCP Interface | MCP Server pattern (6 callable tools) |
| IPFS Gateway | ipfs.io / dweb.link / cloudflare-ipfs |
| MCS Simulation | FilSwan MCS API simulation |
| Data Processing | pandas, numpy |
| Anomaly Detection | Z-score, IQR methods |
| Storage | SQLite (audit trail) |
| Deployment | Vercel |

## 🔍 What's Real vs Mocked

| Component | Status |
|-----------|--------|
| IPFS dataset retrieval | ✅ Real (public IPFS gateways) |
| Data computation (stats, anomalies) | ✅ Real (pandas/numpy) |
| MCP tool interface | ✅ Real (FastAPI endpoints) |
| Data quality scoring | ✅ Real |
| MCS storage simulation | ⚠️ Mocked (simulates FilSwan MCS API) |
| Filecoin deal-making | ⚠️ Mocked (no FIL tokens needed) |

All mockable components are behind interfaces — swap in real MCS SDK and Filecoin deals when FIL is available.

## 🏷️ Tracks Covered

- **Track 2 — Multi-Chain Storage ($20K):** Results stored via MCS simulation, cross-chain storage gateway integration
- **Track 3 — Computing Over Data ($15K):** Core functionality — computation on data retrieved from Filecoin

## 🔗 Links

- **Live Demo:** https://datacompute-agent.vercel.app
- **GitHub:** https://github.com/0xConsole/datacompute-agent
- **Filecoin Docs:** https://docs.filecoin.io
- **FilSwan MCS:** https://docs.filswan.com/multi-chain-storage
- **IPFS:** https://docs.ipfs.io

## 📄 License

MIT