---
title: VC Deal Flow Scout — DeepSeek Edition
emoji: 📈
colorFrom: indigo
colorTo: purple
sdk: gradio
sdk_version: 5.40.0
app_file: app.py
pinned: false
license: mit
short_description: Find startups accelerating on GitHub before they raise
tags:
  - venture-capital
  - github
  - alternative-data
  - deepseek
  - mcp
  - agent
---

# VC Deal Flow Scout — DeepSeek Edition

[![Built with @gitdealflow/mcp-signal](https://signals.gitdealflow.com/api/badge/built-with/svg?variant=long)](https://signals.gitdealflow.com/built-with)

A demo of [DeepSeek-V3](https://huggingface.co/deepseek-ai/DeepSeek-V3) plugged into the public [GitDealFlow](https://signals.gitdealflow.com) MCP server. Ask any question about startup engineering acceleration on GitHub — the agent calls live tools and returns cited answers.

> **Why DeepSeek?** Open-weight + open-MCP = a fully open-source VC research stack. No proprietary inference, no closed data, no tracking. The same agent loop runs locally with `deepseek-v3` self-hosted.

## What you can ask

- "What AI/ML startups are accelerating fastest this month?"
- "Compare engineering momentum at three competing infra companies"
- "Find dark horses in fintech with breakout signals but low public visibility"
- "Show me security startups whose contributor count grew 2x in 90 days"
- "Explain the GitDealFlow methodology"

## Tools available to the agent

| Tool | What it does |
|---|---|
| `get_trending_startups` | Top 20 startups across all sectors by commit-velocity acceleration |
| `search_startups_by_sector` | Drill into one of 20 sectors |
| `get_startup_signal` | Full profile by startup name |
| `get_signals_summary` | Dataset snapshot + citation |
| `get_scout_receipts` | Scout Score 0-100 for any GitHub username |
| `get_methodology` | Full methodology document |

All tools are **read-only**, **idempotent**, and **public** (no auth required for read access). The MCP server is also reachable directly at `https://signals.gitdealflow.com/api/mcp/rpc`.

## How this Space works

1. **DeepSeek-V3** runs on Hugging Face Inference Providers (Together AI / Fireworks fallback)
2. **GitDealFlow MCP server** is reached over HTTPS at `signals.gitdealflow.com/api/mcp/rpc`
3. **Gradio UI** wraps the chat loop and renders the agent's tool calls

The agent loop is ~80 lines of Python. See [`app.py`](./app.py).

## Citation

GitDealFlow data citation:

```
VC Deal Flow Signal (signals.gitdealflow.com), Q2 2026 data.
```

Methodology paper: [SSRN preprint id 6606558](https://ssrn.com/abstract=6606558).

## Local run

```bash
git clone https://huggingface.co/spaces/the-data-nerd/vc-deal-flow-deepseek
cd vc-deal-flow-deepseek
pip install -r requirements.txt
HF_TOKEN=hf_xxx python app.py
```

## Related

- **Companion data explorer (Space):** [the-data-nerd/vc-deal-flow-explorer](https://huggingface.co/spaces/the-data-nerd/vc-deal-flow-explorer) — interactive Gradio dashboard (heatmap + filterable rankings + per-startup drilldown) on top of the same dataset
- **Underlying dataset:** [the-data-nerd/vc-deal-flow-signal](https://huggingface.co/datasets/the-data-nerd/vc-deal-flow-signal) (CC-BY-4.0)
- Public website: https://gitdealflow.com
- npm MCP server: `@gitdealflow/mcp-signal`
- GitHub repo: https://github.com/kindrat86/mcp-deal-flow-signal
- Anthropic Desktop Extensions Directory: submitted 2026-05-02
- Cline MCP Marketplace: [#1491](https://github.com/cline/mcp-marketplace/issues/1491)
- CrewAI Tools: [PR #5682](https://github.com/crewAIInc/crewAI/pull/5682)
- Glama.ai: A-tier, 4.9 TDQS

## License

MIT — see `LICENSE`.