io.github.dan24ou-cpu/agent-signal

AgentSignal

The collective intelligence layer for AI shopping agents.

Every agent that connects makes every other agent smarter. 1,200+ shopping sessions, 95 products, 50 merchants, 10 categories — and growing.

Why this exists: When AI agents shop for users, each agent starts from zero. AgentSignal pools decision signals across all agents so every session benefits from what every other agent has already learned — selection rates, rejection patterns, price intelligence, merchant reliability, and proven constraint matches.

Quick Start (30 seconds)

Remote — zero install, instant intelligence:

{
  "mcpServers": {
    "agent-signal": {
      "url": "https://agent-signal-production.up.railway.app/mcp"
    }
  }
}

Local via npx:

npx agent-signal

Claude Desktop / Claude Code:

{
  "mcpServers": {
    "agent-signal": {
      "command": "npx",
      "args": ["agent-signal"]
    }
  }
}

One Call to Start Shopping Smarter

The smart_shopping_session tool logs your session AND returns all available intelligence in a single call:

smart_shopping_session({
  raw_query: "lightweight running shoes with good cushioning",
  category: "footwear/running",
  budget_max: 200,
  constraints: ["lightweight", "cushioned"]
})

Returns:

• Your session ID for subsequent logging
• Top picks from other agents in that category
• What constraints and factors mattered most
• How similar sessions ended (purchased vs abandoned)
• Network-wide stats

23 MCP Tools

Smart Combo Tools (recommended)

Buyer Intelligence — Shop Smarter

Seller Intelligence — Understand Your Market

Discovery & Monitoring

Write Tools — Contribute Back

Example: Full Agent Workflow

# 1. Start smart — one call gets you session ID + intelligence
smart_shopping_session(category: "electronics/headphones", constraints: ["noise-cancelling", "wireless"], budget_max: 400)

# 2. Evaluate products — get intel as you log
evaluate_and_compare(session_id: "...", product_id: "sony-wh1000xm5", price_at_time: 349, disposition: "selected")
evaluate_and_compare(session_id: "...", product_id: "bose-qc45", price_at_time: 279, disposition: "rejected", rejection_reason: "inferior ANC")

# 3. Compare and close
log_comparison(products_compared: ["sony-wh1000xm5", "bose-qc45"], winner: "sony-wh1000xm5", deciding_factor: "noise cancellation quality")
log_outcome(session_id: "...", outcome_type: "purchased", product_chosen_id: "sony-wh1000xm5")

Every step feeds the network. The next agent shopping for headphones benefits from your data.

Example: Seller Intelligence Workflow

# 1. How is my product performing vs competitors?
get_competitive_landscape(product_id: "sony-wh1000xm5")
# → Category rank #1, 68% head-to-head win rate, beats bose-qc45 on ANC quality

# 2. Why are agents rejecting my product?
get_rejection_analysis(product_id: "bose-qc45")
# → 45% rejected for "inferior ANC", agents chose sony-wh1000xm5 instead 3x more

# 3. What do agents want in my category?
get_category_demand(category: "electronics/headphones")
# → Top demands: noise-cancelling (89%), wireless (82%), unmet need: "spatial audio"

# 4. How does my store perform?
get_merchant_scorecard(merchant_id: "amazon")
# → 34% selection rate, 2% out-of-stock, cheapest option 41% of the time

Categories with Active Intelligence

Agent Framework Examples

Ready-to-run examples in /examples:

All examples connect to the hosted MCP endpoint — no setup beyond pip install required.

REST API

Merchant-facing analytics at https://agent-signal-production.up.railway.app/api:

Self-Hosting

git clone https://github.com/dan24ou-cpu/agent-signal.git
cd agent-signal
npm install
cp .env.example .env  # set DATABASE_URL to your PostgreSQL
npm run migrate
npm run seed           # optional: sample data
npm run dev            # starts API + MCP server on port 3100

Architecture

• MCP Server — Stdio transport (local) + Streamable HTTP (remote)
• REST API — Express on the same port
• Database — PostgreSQL (Neon-compatible)
• 23 MCP tools — 17 read (buyer + seller + discovery) + 6 write

License

MIT