Stealthee MCP - Tools for being early

Stealthee is a dev-first system for surfacing pre-public product signals - before they trend. Built for CTOs and tech leaders who need competitive intelligence and early threat detection. It combines search, extraction, scoring, and alerting into a plug-and-play pipeline you can integrate into Claude, LangGraph, Smithery, or your own AI stack via MCP.

Perfect for competitive intelligence, technology trend monitoring, and strategic planning.

Use it if you're:

• A CTO or tech leader needing competitive intelligence, early threat detection, and innovation scouting to inform strategic decisions
• An investor hunting for pre-traction signals
• A founder scanning for competitors before launch
• A researcher tracking emerging markets
• A developer building agents, dashboards, or alerting tools that need fresh product intel.

What's cookin'?

MCP Tools

Installation & Setup

Prerequisites

• API keys for external services (see Environment Variables)

Quick Start

1. Clone and Setup

   bash

   Copy

   git clone https://github.com/rainbowgore/Stealthee-MCP-tools
   cd stealthee-MCP-tools
   python3 -m venv .venv
   source .venv/bin/activate
   pip install -r requirements.txt
   

2. Configure Environment

   Fill the .env file with your API keys:

   bash

   Copy

   # Required
   TAVILY_API_KEY=your_tavily_key_here
   OPENAI_API_KEY=your_openai_key_here
   NIMBLE_API_KEY=your_nimble_key_here
   
   # Optional
   SLACK_WEBHOOK_URL=your_slack_webhook_here
   

3. Start Servers

   bash

   Copy

   # MCP Server (for Claude Desktop)
   python mcp_server_stdio.py
   
   # FastMCP Server (for Smithery)
   smithery dev
   
   # FastAPI Server (Optional - Legacy)
   python start_fastapi.py
   

Smithery & Claude Desktop Integration

All MCP tools listed above are available out-of-the-box in Smithery. Smithery is a visual agent and workflow builder for AI tools, letting you chain, test, and orchestrate these tools with no code.

Available Tools

• web_search: Search the web for stealth launches using Tavily.
• url_extract: Extract and clean content from any URL.
• score_signal: Use OpenAI to score a single signal for stealthiness.
• batch_score_signals: Score multiple signals in one go.
• search_tech_sites: Search only trusted tech news sources.
• parse_fields: Extract structured fields (like pricing, changelog) from HTML.
• run_pipeline: End-to-end pipeline: search, extract, parse, score, and store.

How to Use in Smithery

1. Open the Stealthee MCP Tools page on Smithery.
2. Click "Try in Playground" to test any tool interactively.
3. Use the visual workflow builder to chain tools together (e.g., search → extract → score).
4. Integrate with Claude Desktop or your own agents by copying the workflow or using the API endpoints provided by Smithery.

Cursor (Stealth Radar MCP)

To use Stealth Radar MCP in Cursor via the hosted URL (Streamable HTTP):

1. Open Cursor Settings → MCP (or search for "MCP" in settings).
2. Under Install MCP Server, fill in:
   • Name: Stealth Radar (or any name you like).
   • Type: streamableHttp.
   • URL: Use either:
     • Smithery: The connection URL from your server's Smithery Connect page (e.g. https://smithery.ai/server/rainbowgore/Product-Stealth-Launch-Radar), or
     • Direct: Your server's MCP endpoint, e.g. https://your-ngrok-url.ngrok-free.app/mcp (must end with /mcp).
3. Click Install. Cursor will connect to the server; once added, it loads automatically when you use Cursor.

If you run the server locally, use stdio instead: set Type to stdio, Command to your Python path, and Args to mcp_server_stdio.py with cwd pointing at the repo.

Claude Desktop Integration

Add to your Claude Desktop config.json file:

{
  "mcpServers": {
    "stealth-mcp": {
      "command": "/path/to/stealthee-MCP-tools/.venv/bin/python",
      "args": ["/path/to/stealthee-MCP-tools/mcp_server_stdio.py"],
      "cwd": "/path/to/stealthee-MCP-tools",
      "env": {
        "TAVILY_API_KEY": "your_tavily_key",
        "OPENAI_API_KEY": "your_openai_key"
      }
    }
  }
}

Tool Use Cases

For Analysts & Builders:

• web_search: Find stealth product mentions across the web
• url_extract: Pull and clean raw text from landing pages
• score_signal: Judge how likely a change log implies launch
• batch_score_signals: Quickly triage dozens of scraped URLs
• search_tech_sites: Limit queries to trusted domains only
• parse_fields: Extract pricing/release info from messy HTML
• run_pipeline: Full pipeline — search → extract → parse → score

Signal Intelligence Workflow

1. Search Phase: Use web_search or search_tech_sites to find relevant URLs
2. Extraction Phase: Use url_extract to get clean content from URLs
3. Parsing Phase: Use parse_fields to extract structured data (pricing, changelog, etc.)
4. Analysis Phase: Use score_signal or batch_score_signals for AI-powered analysis
5. Storage Phase: All signals are stored in SQLite database
6. Alert Phase: High-confidence signals trigger Slack notifications

FastAPI Server

You can also run this project as a FastAPI server for REST-style access to all MCP tools.

Base Endpoints

• Swagger UI: http://localhost:8000/docs
• Health Check: http://localhost:8000/health
• Tool Manifest: http://localhost:8000/tools

Example Usage

Search for stealth launches:

curl -X POST "http://localhost:8000/tools/web_search" \
  -H "Content-Type: application/json" \
  -d '{"query": "stealth startup AI", "num_results": 5}'

Run full detection pipeline:

curl -X POST "http://localhost:8000/tools/run_pipeline" \
  -H "Content-Type: application/json" \
  -d '{"query": "new AI product launch", "num_results": 3}'

Pipeline Parameters

• query (required): Search phrase (e.g. "AI roadmap")
• num_results (optional, default: 5): Number of search results to analyze
• target_fields (optional, default: ["pricing", "changelog"]): Fields to extract from HTML

What run_pipeline Does

1. Searches tech and stealth-friendly sources using Tavily
2. Extracts raw content from each result
3. Parses structured signals (pricing, changelog, etc.)
4. Scores each result with OpenAI to estimate stealthiness
5. Stores results in local SQLite
6. Notifies via Slack if confidence is high

AI Scoring Logic

The score_signal and batch_score_signals tools use GPT-3.5 to evaluate:

• Stealth indicators (e.g. private changelogs, missing press, beta flags)
• Confidence level (Low / Medium / High)
• Textual reasoning (used in UI or alerting)

Database Schema (data/signals.db)

Dev Quickstart (FastAPI)

python start_fastapi.py

Then visit: http://localhost:8000/docs

Built with 💜 for those who spot what others miss.