# MCI Scout

Free public-metadata review over ordinary HTTPS JSON. No account, wallet, MCP server, or x402 payment endpoint is provided.

- [Constraint comparison v2 OpenAPI](https://mci-encounter-lab.az01.chatgpt.site/v2/openapi.json)
- [Combined v1 and v2 OpenAPI description](https://mci-encounter-lab.az01.chatgpt.site/openapi.json)
- [20-request web-search cost example](https://mci-encounter-lab.az01.chatgpt.site/v1/scout/examples/web-search?requests=20)
- [Agent guide and optional consented feedback](https://mci-encounter-lab.az01.chatgpt.site/agent-guide.json)
- [Catalogue scope and observation time](https://mci-encounter-lab.az01.chatgpt.site/v1/catalog/stats)
- [Web form](https://mci-encounter-lab.az01.chatgpt.site/)
- [MCI icon](https://mci-encounter-lab.az01.chatgpt.site/favicon.svg)

For questions about a selected offer, use POST /agent/ask with question offer_facts, cost_and_fit or evidence_limits and one exact resource from /agent-guide.json. All eight offer pages are linked from /offers, with HTML and Markdown. Unsupported questions remain out_of_scope. This route stores no conversation and never executes a provider search. Page-local WebMCP is optional and feature-detected; native discovery/use remains unverified.

For actual constraint checks, POST /v2/scout/compare with schema_version "2". It evaluates declared input/output contracts, request quantity, result limits, payment rail/network/asset, source freshness, quality-evidence requirements, and an explicit budget scope. Eight Tavily/Exa route/mode offers are covered. Responses separate eligible, review and rejected with supported/mismatch/unknown reasons. Eligibility is limited to the requested declared constraints, not execution, independent quality, account/wallet readiness or full cash needed to start. This is not full MCI or an evidence-credit mechanism. V2 may fetch fixed official documentation when its one-hour evidence window expires. Only exact reviewed full-body SHA-256 identity renews evidence; changed, unavailable or unreviewed sources remain unknown. Request inputs are never forwarded. This checks documentation identity, not provider execution or success.

Example for 20 JSON searches and an incremental USD 0.15 list-price budget (explicitly allowing declared-only quality evidence):

POST https://mci-encounter-lab.az01.chatgpt.site/v2/scout/compare
Content-Type: application/json

{"schema_version":"2","capability":"web_search","input_type":"text_query","workload":{"unit":"request","quantity":20},"output":{"format":"json","required_fields":["results[].url"]},"max_results_per_request":10,"ai_summary":false,"payment":{"rail":"provider_account"},"budget":{"scope":"incremental_list_price","currency":"USD","max_amount":"0.15"},"quality":{"require_observed_success":false}}

Omit quality or require observed success to keep untested offers in review. total_cash_outlay budgets remain review-only where minimum top-up, balance, commitments, fees and taxes are unknown. Native x402 USDC atomic prices are separate source-backed offers; USD is never converted to USDC. Requests stay stateless and do not execute or pay providers.

The example uses published USD/credit prices under stated workload assumptions. It is not a live quote, evidence of equal quality, or a verified blockchain payment route. Provider metadata is unverified; no provider is executed. Stateless Scout preview requests do not create a feedback session or get stored by this application. A separately opted-in receipt flow can retain one public six-field request and its complete issued answer for up to 48 hours. Receipt acknowledgment is optional and occurs only after the answer; it does not verify independent external identity or provider use. Only public request counts belong in example URLs; hosting systems may process URL metadata. Feedback is optional and requires separate explicit consent.