# Priorflow Intelligent Agent/Service Selector

> Priorflow is a hosted, agent-native selection service at agentopt.app: give it a natural-language task and it ranks curated MCP servers and A2A agents for that task under real constraints (reliability, cost, latency, autonomy, maturity) using hybrid structured + semantic matching over a catalog of about 5,000 active / 7,900 total candidates. It ranks; it does not invoke the tools it recommends, it is not an MCP server, and it does not run the A2A task runtime. The whole product is one call: POST https://agentopt.app/v1/select.

This file was generated by API Evangelist from the provider's published surface on 2026-09-19; Priorflow does not publish its own /llms.txt (probed, 404). Operator: Vitaly Kantorovich, per the agent card's provider.organization. No OpenAPI, SDK, terms of service or privacy policy is published.

## Machine surface

- [A2A agent card](https://agentopt.app/.well-known/agent-card.json): protocolVersion "1.0", version 0.2.0, two skills (select-agents, describe-catalog). Also served byte-identically at /.well-known/agent.json. Its top-level `priorflow` block is the only machine-readable contract for the API: request shape, tier field lists, score scales, 402/429 semantics and the key-purchase flow.
- [Select endpoint](https://agentopt.app/v1/select): POST, JSON body `{query, top_n?, constraints?, weights?, caller_agent?}` (plus paid-only `include_explanations`). Returns ordered candidates with an absolute 0-1 fit score (0.55 usable, 0.72+ strong). Free tier: id, score, score_band, name, top_n <= 5. Paid tier (X-API-Key or Authorization: Bearer): adds source_url, endpoint/connect, homepage_url, endpoint_status, tags, dimensions, recommendation, clarifications, score_breakdown; top_n <= 20.
- [Readiness](https://agentopt.app/ready) and [health](https://agentopt.app/health): JSON. /ready reports catalog size, require_api_key, publish_agent_card, mcp_enabled (false) and a2a_runtime_enabled (false).
- robots.txt disallows /v1/, /mcp and /a2a to crawlers; agents are expected to call /v1/select directly, not crawl it.

## Docs

- [Info](https://agentopt.app/info): the product page and the only documentation — copy-paste curl for free and paid calls, tiers, matching dimensions, score bands, and what the service is and is not. Identical to the site root.
- [Try](https://agentopt.app/try): browser form over the production /v1/select for human exploration; an operator-published shared evaluation key with a stated expiry appears here.
- [Upgrade](https://agentopt.app/upgrade): how paid keys are obtained — Stripe Checkout at POST /v1/billing/checkout, one-time reveal at GET /v1/billing/session/{session_id}/key; 401/402/429 semantics.
- [Status](https://agentopt.app/status): daily Observer scorecard — catalog counts, synthetic select success and relevance rates, latency bands, and a dated trend table. Machine JSON at /v1/status.
- [Explore](https://agentopt.app/explore): JS-rendered catalog map (domains, quality bands, modalities, skill density).

## How to call it

- Free tier: no key. Send `Content-Type: application/json`, a non-empty `User-Agent`, and a stable `caller_agent` in the body. Volume is roughly 20/min, 120/hour, 500/day per IP; exhaustion is a 429 whose JSON body may carry a machine-readable `upgrade` object.
- Paid tier: send the `pf_live_` key as `X-API-Key` (or `Authorization: Bearer`). Invalid keys always 401. Paid features without a key return 402 `upgrade_required`; an exhausted pack returns 402 `select_quota_exceeded`; a lapsed time-limited key returns 402 `api_key_expired`.
- Scores are absolute, not relative: 0.55 is a usable first pick, 0.72+ a strong one. Dimension ratings are 0-1 with higher always better (cost higher = cheaper; latency higher = faster).
- Priorflow returns candidates for the caller to choose from; it does not proxy or invoke them.

## API Evangelist artifacts

- [apis.yml](https://raw.githubusercontent.com/api-evangelist/agentopt-app/refs/heads/main/apis.yml)
- [A2A manifest (graded)](https://raw.githubusercontent.com/api-evangelist/agentopt-app/refs/heads/main/a2a/agentopt-app-a2a.yml)
- [Agent card (verbatim copy)](https://raw.githubusercontent.com/api-evangelist/agentopt-app/refs/heads/main/a2a/agentopt-app-agent-card.json)
- [Authentication](https://raw.githubusercontent.com/api-evangelist/agentopt-app/refs/heads/main/authentication/agentopt-app-authentication.yml)
- [Conventions](https://raw.githubusercontent.com/api-evangelist/agentopt-app/refs/heads/main/conventions/agentopt-app-conventions.yml)
- [Error catalog](https://raw.githubusercontent.com/api-evangelist/agentopt-app/refs/heads/main/errors/agentopt-app-problem-types.yml)
- [Rate limits](https://raw.githubusercontent.com/api-evangelist/agentopt-app/refs/heads/main/rate-limits/agentopt-app-rate-limits.yml)
- [Plans](https://raw.githubusercontent.com/api-evangelist/agentopt-app/refs/heads/main/plans/agentopt-app-plans-pricing.yml)
- [Well-known probe](https://raw.githubusercontent.com/api-evangelist/agentopt-app/refs/heads/main/well-known/agentopt-app-well-known.yml)