# 🚀 prelaunch-mcp

**Smarter pre-build reality check for AI agents.**

Stop building what already exists. `prelaunch-mcp` scans 6 sources with LLM-powered intent parsing to tell you if your idea has competition, if there's real demand, and where the gaps are.

## ✨ What Makes This Different

| Feature                  | Other validation tools       | **prelaunch-mcp**                                   |
| ------------------------ | ---------------------------- | --------------------------------------------------- |
| **Intent understanding** | Dictionary keyword stripping | 🧠 LLM-powered intent parsing                       |
| **Search queries**       | Generic keyword variants     | Targeted competitor-finding queries                 |
| **Sources**              | 3-5 dev-only sources         | **6** (GitHub, HN, npm, PyPI, Google/DDG, Reddit)   |
| **Scoring**              | Simple bucket math           | Relevance-weighted with intelligent caps            |
| **Demand signals**       | ❌ None                      | ✅ Reddit pain signal detection                     |
| **Insights**             | Static templates             | Context-aware dynamic insights                      |
| **Output**               | Single score                 | `competition_score` + `demand_score` + gap analysis |

## 📦 Installation

### For Claude Code / Cursor / Windsurf

```bash
# Add to your MCP config
claude mcp add prelaunch -- uvx prelaunch-mcp
```

### For other MCP clients

Add to your MCP config (e.g., `.claude.json`, `mcp.json`):

```json
{
  "mcpServers": {
    "prelaunch": {
      "command": "uvx",
      "args": ["prelaunch-mcp"],
      "env": {
        "ANTHROPIC_API_KEY": "your-key-here",
        "GITHUB_TOKEN": "your-token-here"
      }
    }
  }
}
```

### Run directly

```bash
uvx prelaunch-mcp
```

## 🔑 Environment Variables

**All API keys are user-provided. We never store or transmit your keys.**

| Variable            | Required    | Purpose                                           |
| ------------------- | ----------- | ------------------------------------------------- |
| `ANTHROPIC_API_KEY` | Recommended | LLM intent parsing (Claude Haiku — ~$0.001/check) |
| `OPENAI_API_KEY`    | Alternative | LLM parsing via OpenAI or compatible API          |
| `OPENAI_BASE_URL`   | Optional    | Custom endpoint (Ollama, LM Studio, etc.)         |
| `GITHUB_TOKEN`      | Optional    | Higher GitHub API rate limits                     |
| `GOOGLE_CSE_KEY`    | Optional    | Google Custom Search (falls back to DuckDuckGo)   |
| `GOOGLE_CSE_ID`     | Optional    | Google Custom Search Engine ID                    |

**Without any API keys**, the tool still works using fallback keyword extraction and all free sources (GitHub, HN, PyPI, npm, DuckDuckGo, Reddit).

## 🎯 Usage

Once installed, your AI agent can call it naturally:

```
"Check if anyone has built an AI agent security scanner"
"Is there competition for a Kubernetes cost optimization dashboard?"
"Run a pre-launch check on: open-source compliance tool for Indian banks"
```

### Depth Modes

| Mode       | Speed       | Sources               | LLM                    |
| ---------- | ----------- | --------------------- | ---------------------- |
| `quick`    | ⚡ Fast     | GitHub + HN           | No                     |
| `standard` | 🔄 Balanced | All 6 sources         | Yes (if key available) |
| `deep`     | 🔍 Thorough | All 6 + extra queries | Yes (if key available) |

## 📊 Example Output

```json
{
  "competition_score": 42,
  "demand_score": 60,
  "competition_level": "high",
  "intent": {
    "category": "security",
    "product_type": "CLI tool",
    "target_audience": "AI engineers",
    "target_technology": "LangChain, CrewAI",
    "analogy": "npm audit but for AI agents",
    "core_problem": "No automated security scanning for AI agent deployments"
  },
  "top_similars": [...],
  "pain_signals": [
    {
      "title": "Anyone built security tooling for LangChain agents?",
      "url": "https://reddit.com/...",
      "subreddit": "LangChain",
      "score": 47
    }
  ],
  "insights": [
    "🟡 Moderate competition — the space exists but isn't saturated.",
    "🔥 Strong demand signal: 3 Reddit posts expressing unmet need.",
    "✅ Promising: competition validates market AND demand signals confirm unmet needs.",
    "🧭 Analogy: \"npm audit but for AI agents\" — validate dynamics apply to security."
  ]
}
```

## 🔧 Development

```bash
# Clone
git clone https://github.com/Heman10x-NGU/prelaunch-mcp.git
cd prelaunch-mcp

# Install deps
uv sync --dev

# Run tests
uv run pytest tests/ -v

# Run server locally
uv run prelaunch-mcp
```

## Architecture

```
Input: "AI agent security scanner like npm audit for CrewAI"
    │
    ├── Stage 1: LLM Intent Parse (or fallback keywords)
    │   → category: "security" | type: "CLI tool" | queries: [...]
    │
    ├── Stage 2: Multi-Source Scan (parallel)
    │   → GitHub, HN, PyPI, npm, Google/DDG, Reddit
    │
    └── Stage 3: Scoring & Analysis
        → competition_score + demand_score + gap insights
```

## License

MIT