cited-answer

name

cited-answer

description

Answer a research / data question by discovering and BUYING exactly the paid data needed over L402 (per-call, no account), then writing an answer with citations and a satoshi receipt. Use when the user wants a factual, sourced answer that benefits from real paid data (market figures, economic series, weather, etc.) and is willing to spend a small amount of sats. Do NOT use for questions answerable from general knowledge or free sources.

Cited answer

Turn a question into a sourced answer paid for per-call over L402 — discover the data, buy only what you need, cite every figure, and show the receipt. This is agentic commerce as a research tool: the agent pays its own way for exactly the inputs it needs, account-lessly.

What you need

• Lightning Enable MCP with a funded wallet (discover_api, access_l402_resource, get_budget_status, get_payment_history).
• A small sats budget the user is comfortable spending.

Flow

1. Clarify the question enough to know what data would actually support it. Decide which figures need real, current data vs. general knowledge.

2. Set a budget ceiling with the user up front (e.g. "up to 25 sats"). Check get_budget_status so you know your headroom before spending anything.

3. Discover sources with discover_api (search the L402 registry, e.g. discover_api(query="...")). With budget_aware on, it annotates how many calls you can afford. Prefer the cheapest source that has the data.

4. Buy only what you need, one call at a time. Use access_l402_resource on each endpoint — it runs the full 402 → pay → retry cycle and returns the data. Track what you spent on each call (source + sats + what it bought).

5. Stay under budget. Re-check get_budget_status as you go; stop before exceeding the ceiling and tell the user if you need more.

6. Write the answer grounded in the purchased data. Every figure must trace to a specific paid call. If the real data contradicts the obvious narrative, say so — that's the value of paying for real data over guessing.

7. Show the receipt. Build a JSON array of your paid calls and render it:

   echo '[{"source":"FRED USD/JPY","sats":5,"what":"DEXJPUS series"}, ...]' \
     | python scripts/receipt.py
   

   Present the cited answer, then the receipt (total sats, calls, 0 accounts).

Safety / honesty rules

• Respect the budget ceiling. Never blow past it; ask first.
• Cite honestly. Only claim a figure if a paid call actually returned it. Don't dress up a guess as sourced.
• Cheapest-sufficient source wins — don't over-buy.
• If no affordable source has the data, say so plainly rather than inventing it.

Example

User: "Write me a sourced take on the 2024 yen carry-trade unwind. Up to 25 sats."

→ get_budget_status → ok. discover_api(query="finance") → finds FRED on agent-commerce.store. → access_l402_resource for USD/JPY, Nikkei, VIX, 10-yr (5 sats each). → write the analysis, every number traced to a call, noting the data distinguishes the Aug-2024 unwind from the bigger Apr-2025 shock. → receipt.py → "20 sats, 4 calls, 0 accounts, 0 API keys."