# AgentSort

> AgentSort is the capability retrieval engine for AI. Given an intent, it retrieves, verifies and ranks the small set of trusted capabilities a frontier model should consider — across MCP, A2A, ARD, APIs, Agent Skills and future protocols.

Primary statement: The capability retrieval engine for AI.
Strategic statement: The ranking layer for the agentic internet.
Developer statement: One connection. Every agent.
Frontier-model proposition: We retrieve the candidates. Your model keeps the judgment.
Business proposition: Make sure AI can find you, choose you and transact with you.
Website: https://www.agentsort.com

## The problem
An AI with 20 tools can reason about them directly. An AI with access to 20 million capabilities cannot.
As MCP servers, A2A agents, ARD catalogs, APIs, Agent Skills and machine-actionable businesses grow from
thousands to millions, no frontier model can permanently load or reason over every possible tool.

## The flow
Intent → Retrieve → Rank → Verify → Return candidates → Frontier model chooses → Execute → Outcome → AgentSort learns.

## Products
- **AgentSort Search** — what exists?
- **AgentRank** — which candidates are best for this task? Organic ranking is never for sale.
- **AgentSort Resolve** — which small set of capabilities (typically 3–7) should this AI consider right now?
- **AgentSort Gateway / Route** — execute through the selected capability.
- **Observe / Execution Graph** — measure what actually happened, across platforms.
- **AgentVerify** — four verification levels: identity, capability, security, execution.
- **AgentSort Ready** — makes a business discoverable and transactable by AI (ARD, MCP, A2A cards, OpenAPI, availability, pricing).
- **AgentExchange** — a separate, clearly labelled commercial feed; evolves from cost per invocation toward cost per outcome (CPO).
- **AgentSort Enterprise** — discover, policy filter, verify, rank, route and audit internal agents and tools.

## The five graphs
Business Graph (who exists) → Capability Graph (what they can do) → Access Graph (how an AI reaches them)
→ Commercial Graph (cost, availability, inventory) → Execution Graph (what actually happened) → Outcome.

## Principles
- Organic AgentRank is never for sale; AgentExchange is a separate labelled feed.
- AgentSort returns candidates and evidence; the frontier model keeps final judgment.
- All statistics and demos on the site are illustrative until real production data exists.
- No formal partnerships with any frontier-model provider are claimed.

## Pages
- [Home](https://www.agentsort.com/): capability retrieval thesis, Resolve demo, verification ladder, execution graph.
- [Platform](https://www.agentsort.com/platform): products, ranking signals, verification, the AgentSort Graph.
- [Developers](https://www.agentsort.com/developers): APIs, MCP endpoint, build-vs-buy, protocol coverage.
- [For Businesses](https://www.agentsort.com/businesses): Agent Visibility Score, AgentSort Ready, AgentExchange.
- [Company](https://www.agentsort.com/company): why now, the search-engine analogy, the learning flywheel.