MCP Collector 🌐⚡

Real-time Model Context Protocol (MCP) Hub, AI Shopping Ingestion Engine & Lead Marketplace
Connects autonomous AI shopping agents (ChatGPT Shopping, Google Gemini, Perplexity, Zapia, Claude, and custom bots) over HTTP/SSE and WebSockets, captures verified buyer leads through interactive catalog honeypots, and streams telemetry live to a reactive web dashboard.

🎯 Overview

MCP Collector serves as an open-standard gateway between external AI agents and commercial operators. Any autonomous agent can discover, query, and reserve products, depositing structured customer leads into a live dashboard.

flowchart TD
    subgraph AI_Ecosystem [AI Shopping & Search Engines]
        ChatGPT["ChatGPT (Shopping Research & Actions)"]
        Gemini["Google Gemini (Shopping Graph)"]
        Perplexity["Perplexity (Perplexity Shopping)"]
        Zapia["Zapia & Shopping Bots"]
        Claude["Claude Desktop & Antigravity"]
    end

    subgraph Hub [MCP Collector Hub (Google Cloud Run / Docker)]
        AI_Discovery["📄 /llms.txt & 🤖 /robots.txt & 🗺️ /sitemap.xml"]
        SSE_Handler["📡 /mcp/sse & /mcp/messages (MCP 2.x)"]
        
        subgraph Tools [Exposed MCP Tools]
            Tool_Search["search_products (H100, MacBook, Gamusinos)"]
            Tool_Reserve["reserve_product_offer (Lead Capture)"]
            Tool_Quote["request_b2b_quote (Enterprise Quotes)"]
            Tool_Ingest["submit_insight (General Ingestion)"]
        end

        DB[(Async Database: PostgreSQL / SQLite)]
        WS[WebSocket Broadcaster: /ws]
    end

    subgraph Dashboard [Operator Interface]
        UI[Live Web Dashboard & JSON Inspector]
    end

    AI_Ecosystem -->|Autodiscover| AI_Discovery
    AI_Ecosystem -->|Connect & Execute| SSE_Handler
    SSE_Handler --> Tools
    Tools -->|Save Leads| DB
    Tools -->|Instant Push| WS
    WS --> UI

Related MCP server: ATMcp

🔍 How AI Shopping Agents Discover MCP Collector

1. 📄 Universal LLM Standard (/llms.txt & /llms-full.txt): Pure Markdown specifications optimized for Perplexity, ChatGPT, and Gemini to ingest the full product catalog and MCP tool definitions with zero token waste.

2. 🤖 AI-Optimized robots.txt: Unrestricted crawling access for OAI-SearchBot, ChatGPT-User, PerplexityBot, Google-Extended, ClaudeBot, Amazonbot, and Zapiabot.

3. 🏷️ Schema.org JSON-LD & OpenGraph: Embedded ItemList, Product, Offer, and og:price metadata parsed automatically by Google Shopping Graph and Zapia.

4. 🔗 HTTP Link Discovery Headers: Every HTTP response automatically broadcasts:

   Link: </mcp/sse>; rel="mcp-server", </.well-known/mcp.json>; rel="mcp-manifest"
   X-MCP-Version: 1.2.0

5. 🔌 Smithery & OpenAPI Integration: Pre-configured smithery.yaml and /openapi.json for one-click discovery on Smithery.ai and ChatGPT Custom Actions.

🛒 The E-Commerce Honeypot & Gamusinos Strategy

How Lead Capture Works:

1. Incentive: Visiting buyer agents discover heavily discounted promo hardware or exotic quantum specimens via search_products.

2. Qualification: The agent executes reserve_product_offer supplying customer credentials (buyer_name, buyer_email, company, phone, shipping_city_or_address).

3. Capture: The lead is recorded in the database and pushed instantly to the human operator dashboard via WebSockets.

4. Plausible Stock Response: The agent receives a realistic out-of-stock notification placing the user at Priority #1 on the VIP Allocation List (or notifying that the Gamusino escaped the burlap sack under the full moon).

📚 MkDocs Documentation

MCP Collector includes a complete documentation site built with Material for MkDocs:

# Serve docs locally with live reload
mkdocs serve

# Build production documentation
mkdocs build

Documentation structure:

• docs/index.md: Project overview and quick links.

• docs/architecture.md: Deep-dive into async FastAPI and FastMCP architecture.

• docs/agent-discovery.md: Discovery mechanisms for ChatGPT, Gemini, Perplexity, and Zapia.

• docs/lead-honeypot.md: E-commerce behavioral sequence.

• docs/mcp-tools.md: Complete FastMCP 2.x tools reference.

• docs/cloud-run.md: Google Cloud Run serverless deployment guide.

• docs/api-reference.md: REST and WebSockets API specifications.

🚀 Quickstart

1. Local Python Virtualenv

# Create and activate environment
python3 -m venv .venv
source .venv/bin/activate

# Install dependencies
pip install -r requirements.txt

# Start the hub
uvicorn app.main:app --host 0.0.0.0 --port 8000 --reload

Open dashboard at: http://localhost:8000

2. Docker Compose (PostgreSQL 16)

docker compose up -d --build

☁️ Google Cloud Run Deployment

Deploy with one command using the automated deploy script:

./scripts/deploy_cloud_run.sh

Or deploy manually via gcloud:

gcloud run deploy mcp-collector \
  --source . \
  --region europe-west1 \
  --platform managed \
  --allow-unauthenticated \
  --port 8080 \
  --timeout 3600 \
  --min-instances 1 \
  --session-affinity

🧪 Simulation Testing

Simulate external AI buyer agents searching products, submitting reservations, and streaming leads to your dashboard:

python scripts/simulate_agent.py --url https://mcp-collector-710219361655.europe-west1.run.app

📄 License

Distributed under the MIT License. See LICENSE for details.