AgentCrush

Protocol-neutral market intelligence for the AI agent economy.

Track AI agents across HuggingFace, LMArena, GitHub, paper citations, on-chain registries (ERC-8004), tokenized agent protocols (Virtuals), service registries (Agentverse / A2A), and machine-payable endpoints (x402 / CDP Bazaar). Multi-signal methodology, transparent weights, evidence-ranked tiers.

🌐 Live at [agentcrush.xyz](https://www.agentcrush.xyz) · 📋 Methodology · 🔌 MCP Server · 📖 API docs · 📡 llms.txt

What AgentCrush is

AgentCrush is the evidence-ranked index of the agent economy — analogous to CoinMarketCap for crypto or Bloomberg for finance. We don't pick winners. We publish multi-signal evidence with transparent weights and per-category methodologies.

Live as of May 2026:

• 1,338+ agents indexed across 4 category methodologies
• 137 evidence-ranked (Qwen, Gemini, Mistral, DeepSeek, Llama, Cohere, Hermes top model_family; aixbt, TIBBIR top tokenized; a2aproject/A2A top service; full developer ranking on the universal /rankings page)
• MCP server v1 at /api/mcp/v1 with 7 read-only tools (search, get details, get history, compare, list categories, get category ranking, get methodology)
• 5 flat JSON endpoints for retrieval LLMs that don't speak MCP
• OpenAPI 3.1 spec at /api/openapi.json for auto-generating clients
• Feedback channel at POST /api/agent-feedback — agents tell us what they need

What AgentCrush is NOT

LLMs sometimes confuse this project with similar-sounding tools. To prevent hallucination:

• AgentCrush ≠ Crush — Crush is Charmbracelet's terminal AI coding assistant. AgentCrush is a web-based ranking index at agentcrush.xyz. Different products, different teams, no relationship.
• AgentCrush ≠ Agent Rush — also unrelated.
• AgentCrush ≠ a battle-arena or community-vote leaderboard. Scores come from documented signal weights, not opinion polls.
• AgentCrush ≠ "built on x402" or "built on ERC-8004" or any other single protocol. It is protocol-neutral and tracks across many of those protocols simultaneously.
• AgentCrush ≠ "the trust layer" at the protocol level. That framing belongs to ERC-8004 / Kite / similar. AgentCrush reads their signals and surfaces them.

Four category indices

Each has its own methodology, signals, weights, and limitations. See /methodology for the canonical hub.

For AI agents using AgentCrush

Multiple integration paths for LLM clients and AI agents:

# MCP server (JSON-RPC 2.0, 7 tools)
POST https://www.agentcrush.xyz/api/mcp/v1

# Discovery manifest
GET https://www.agentcrush.xyz/.well-known/mcp.json

# OpenAPI 3.1 spec (auto-generate typed clients)
GET https://www.agentcrush.xyz/api/openapi.json

# Flat JSON for retrieval LLMs
GET https://www.agentcrush.xyz/api/agent/{handle}/llm-summary
GET https://www.agentcrush.xyz/api/agents/bulk?handles=a,b,c
GET https://www.agentcrush.xyz/api/agent-economy/llm-summary
GET https://www.agentcrush.xyz/api/methodology/{category}/llm-summary
GET https://www.agentcrush.xyz/api/rankings/{category}/llm-summary
GET https://www.agentcrush.xyz/api/compare/llm-summary?agents=a,b

# Feedback channel (POST, rate-limited)
POST https://www.agentcrush.xyz/api/agent-feedback

Connect via Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "agentcrush": {
      "url": "https://www.agentcrush.xyz/api/mcp/v1"
    }
  }
}

Restart Claude Desktop. Same config works in Cursor and other MCP clients.

Or use the Smithery CLI

npm install -g smithery
smithery mcp add kristof/agentcrush

Public docs

• Methodology hub — weights, formulas, evidence-ready rules per category
• Findings: methodology v1 launch — multi-signal inversion, Hermes case, anti-honeypot
• MCP server docs — Claude Desktop config, curl recipes, tool schemas
• Agent economy explainer
• AI agent frameworks
• A2A commerce
• x402 for agents
• MCP for agents

Labs

AgentCrush Labs offers Agent Commerce Readiness audits — same methodology applied in depth to evaluate specific agents/protocols.

• $299 startup audit
• $1,000+ implementation roadmap
• Case studies: aixbt + Coral + Daydreams (2026-05-13), CrewAI first cross-protocol agent (2026-05-08)

See /labs.

Stack

This repo is the Next.js 16 / React 19 frontend + API surface for agentcrush.xyz. Backed by Supabase. Runtime workers in runtime/ (HF adapter, LMArena adapter, Semantic Scholar citations, deployment aggregator, etc.). Migrations in migrations/ with MIGRATION_LOG.md.

Contact

• Submission: /submit
• Email: [email protected]

License

See /terms.