Leads Plaza MCP Server

An MCP server that lets AI agents search, buy, and report on sales leads — with x402 micropayments (USDC on Base) and a conversion-feedback loop that drives QA scoring and dynamic pricing.

5-minute quickstart

git clone https://github.com/YOUR-ORG/leads-plaza-mcp-server.git
cd leads-plaza-mcp-server
python3 -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
python test_stdio.py   # runs a full self-check over real MCP stdio

Then register it with your agent framework.

Claude Code (~/.claude.json or project .mcp.json):

{
  "mcpServers": {
    "leads-plaza": {
      "command": "/path/to/leads-plaza-mcp-server/.venv/bin/python",
      "args": ["server.py"],
      "cwd": "/path/to/leads-plaza-mcp-server",
      "env": {
        "LEADS_NETWORK": "base-sepolia",
        "LEADS_PAY_TO": "0xYourSellerAddress"
      }
    }
  }
}

Cursor (Settings → MCP → Add server, same command/args/env as above).

No LEADS_PAY_TO? The server still runs — buy_lead returns a placeholder 402 explaining that no real payment will happen. Set mock_confirm=true to simulate a paid purchase locally.

Related MCP server: payanagent

Tools

Categories: ai_automation_intent, funding_hiring_signal, devtool_signal.

Network status

Testnet only (October 2026). The x402 flow is live on Base Sepolia (eip155:84532, LEADS_NETWORK=base-sepolia). Mainnet is not open yet — do not send real funds. When mainnet opens, set LEADS_NETWORK=base (default) and a real LEADS_PAY_TO seller address.

Architecture

Agent / framework (Claude Code, Cursor, …)
        │  MCP stdio
        ▼
  server.py  ── search_leads / buy_lead / report_conversion
     │                │                  │
     │                │                  ▼
     │                │           conversions.db (sqlite)
     │                │           orders · conversions · buyer_credits
     │                ▼
     │          x402.py ── 402 paymentRequirements (exact/USDC)
     │                     verify/settle via Coinbase facilitator
     ▼
  pricing.py ── price = base × freshness × qa_coef × cvr × exclusive
  mock_data.py ── 12 fictional seed leads (used when no scored data present)

Data fallback: on startup the server looks for scored QA leads next to the repo (../data/qa/scored_leads_*.jsonl in a full deployment); when absent it gracefully falls back to the bundled mock_data.py fixtures. No real personal data is ever shipped — mock contacts use *.local placeholder domains.

Local verification

python test_stdio.py

Runs list_tools → search → buy (402 placeholder) → buy (mock paid) → report_conversion over the real MCP stdio protocol. All green means the loop works.

License

MIT — see LICENSE.