# Autonomous Venture Capital

> Autonomous AI agents that create and fund startups in real-time. Built with CrewAI, Gemini 2.5 Flash, FastAPI, and Model Context Protocol (MCP).

- **Type:** MCP server
- **Install:** `agentstack add mcp-zenmakhlouf-autonomous-venture-capital`
- **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior
- **Seller:** [zenmakhlouf](https://agentstack.voostack.com/s/zenmakhlouf)
- **Installs:** 0
- **Category:** [Integrations](https://agentstack.voostack.com/c/integrations)
- **Latest version:** 0.1.0
- **License:** MIT
- **Upstream author:** [zenmakhlouf](https://github.com/zenmakhlouf)
- **Source:** https://github.com/zenmakhlouf/autonomous-venture-capital

## Install

```sh
agentstack add mcp-zenmakhlouf-autonomous-venture-capital
```

Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent.

## About

# 🚀 AI Crowdfunding Platform

**Autonomous AI agents that create and fund software startups in real time.**

An end-to-end demonstration of agentic AI development: a **Founder Agent** imagines startups and an **Investor Agent** evaluates and funds them — all streamed live to a React dashboard. Built with **CrewAI + Google Gemini**, **FastAPI**, and the **Model Context Protocol**.

---

## Why I Built This

I wanted to build something that demonstrates autonomous multi-agent collaboration beyond "chatbot with tools." This project showcases:

- **Agents with real autonomy** — the Founder creates projects from scratch; the Investor makes independent funding decisions using a scoring heuristic.
- **Tool-calling without regex** — Gemini's native function-calling eliminates the fragile output-parsing common in agent frameworks.
- **Full-stack observability** — every agent thought, tool call, and result streams to an event bus for real-time visualization.
- **Protocol-driven connectivity** — the backend exposes its capabilities via **MCP**, allowing it to be controlled by any compliant LLM client (like Claude Desktop).

---

## Features

### 1. Autonomous Agent Loop

Agents operate in a continuous loop:

1. **Founder** invents a startup concept, name, and funding goal.
2. **Investor** analyzes the deal flow, scoring projects on innovation and market fit.
3. **Execution**: if a project meets the criteria, the Investor deploys capital instantly.

_Live event stream (right) showing real-time agent reasoning and tool usage._

### 2. High-Fidelity Agent Output

Agents don't just "talk" — they execute complex workflows. Below is the terminal output showing a full cycle where an Investor Agent evaluates projects and commits funding.

### 3. Universal MCP Integration

The entire platform is an MCP server. This means you can connect **Claude Desktop** (or any MCP client) to your running backend and interact with the data naturally.

**"Claude, find me all AI projects seeking under $50k"** -> Claude calls the `search_projects` tool directly.

---

## Architecture

```mermaid
graph TD
    Client[React Dashboard] -->|HTTP/WebSocket| API[FastAPI Backend]

    subgraph "AI Agent Layer (CrewAI)"
        Founder[Founder Agent]
        Investor[Investor Agent]
    end

    subgraph "Capabilities"
        Tools[Tool Definitions]
        LLM[Gemini 2.5 Flash]
    end

    Founder -->|Uses| Tools
    Investor -->|Uses| Tools
    Tools -->|HTTP| API

    MCP[MCP Server] -->|stdio| API
    Claude[Claude Desktop] -->|MCP Protocol| MCP
```

---

## Key Design Decisions

| Decision                           | Rationale                                                                                                                                                |
| ---------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Gemini over local LLMs**         | Gemini 2.5 Flash reliably calls tools without needing regex fallbacks — reduced agent code by ~65%.                                                      |
| **Heuristic + LLM hybrid scoring** | The Investor uses a deterministic scoring formula first, then lets the LLM reason about the top candidates. This makes investment decisions explainable. |
| **Event-driven architecture**      | Every agent thought is a structured event (`agent_thinking`, `tool_call`, `tool_result`), enabling real-time dashboards and audit trails.                |
| **MCP integration**                | The same backend that agents use is also exposed as MCP tools, so any LLM client can interact with the platform without custom glue code.                |
| **Shared LLM factory**             | Single `build_llm()` function replaces duplicated config across agents — robust configuration management.                                                |

---

## Quick Start

### Prerequisites

- Python 3.11+
- Node.js 18+
- A [Gemini API key](https://aistudio.google.com/apikey)

### Setup

```bash
# Clone and enter the project
git clone https://github.com//ai-crowdfund-platform.git
cd ai-crowdfund-platform

# Create a virtual environment
python -m venv .venv
source .venv/bin/activate

# Install Python dependencies
pip install -r requirements.txt

# Configure environment
cp .env.example .env
# Edit .env and add your GEMINI_API_KEY
```

### Run

```bash
# Terminal 1 — Start the backend
uvicorn backend.app:app --reload --port 8000

# Terminal 2 — Start the frontend
cd frontend && npm install && npm run dev

# Terminal 3 — Run the autonomous agents
python -m agents.runner
```

Then open **http://localhost:5173** to see the dashboard.

---

## Connecting context-aware AI (Claude)

To let Claude Desktop control the platform:

1. Add this to your `claude_desktop_config.json`:

```json
{
  "mcpServers": {
    "crowdfund": {
      "command": "/path/to/project/.venv/bin/python",
      "args": ["/path/to/project/mcp_server/run.py"],
      "env": {
        "BACKEND_URL": "http://localhost:8000"
      }
    }
  }
}
```

2. Restart Claude. You can now ask: _"Check the crowdfunding platform for new sports startups."_

---

## Tech Stack

| Layer          | Technology                      |
| -------------- | ------------------------------- |
| **AI Agents**  | CrewAI, Google Gemini 2.5 Flash |
| **Backend**    | FastAPI, Pydantic, WebSocket    |
| **MCP Server** | Model Context Protocol (stdio)  |
| **Frontend**   | React 18, Vite, Tailwind CSS    |
| **Testing**    | pytest, FastAPI TestClient      |

---

## What I'd Add Next

- **Due Diligence Agent** — A third agent that researches market data before the Investor commits capital.
- **Pydantic structured output** — Replace free-text reflection with typed `InvestmentDecision` models.
- **Agent memory** — Persistent memory so the Investor recalls past decisions and avoids over-concentration.
- **Docker Compose** — One-command launch for backend + frontend + agents.

---

## License

MIT

## Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [zenmakhlouf](https://github.com/zenmakhlouf)
- **Source:** [zenmakhlouf/autonomous-venture-capital](https://github.com/zenmakhlouf/autonomous-venture-capital)
- **License:** MIT

Install and usage instructions live in the source repository linked above.

## Pricing

- **Free** — Free

## Security capabilities

Automated source analysis of v0.1.0 — what this tool can access:

- **Network access:** yes
- **Filesystem access:** no
- **Shell / process execution:** no
- **Environment & secrets:** yes
- **Dynamic code execution:** no

*"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.*


## Versions

- **0.1.0** — security scan: passed — Imported from the upstream source.

## Links

- Listing page: https://agentstack.voostack.com/l/mcp-zenmakhlouf-autonomous-venture-capital
- Seller: https://agentstack.voostack.com/s/zenmakhlouf
- Browse the marketplace: https://agentstack.voostack.com/browse

---
Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.