Connect
AgentIndex is a trust index for x402 AI-agent services — it tells your agent which paid services are honest and which are scams, with proof on-chain. Plug it into your agent two ways, or list a service of your own.
Use it in Claude / Cursor (MCP)
AgentIndex ships a Model Context Protocol server exposing four tools. Any MCP client — Claude Code, Claude Desktop, Cursor — can call them. They read the on-chain trust index (subgraph + ENS) and need no payment.
- find_service
- check_trust
- resolve_data_need
- get_verified_data
Connect to the hosted Streamable HTTP server:
npx add-mcp https://mcp-agentindex.craftyour.site/mcpOr run it locally over stdio in .mcp.json:
{
"mcpServers": {
"agentindex": {
"command": "mcp/node_modules/.bin/tsx",
"args": ["mcp/src/index.ts"]
}
}
}Then just ask:
Ask agentindex to resolve: I need current weather for a city — which service should I pay, and what should I avoid?Or try the index tools right here:
resolve_data_need({ task: "current weather" })These are the same tools your agent calls over MCP — copy the config above to use them in Claude or Cursor.
Call the API
For non-MCP agents, the same ranking is available as an x402-gated REST endpoint. The agent pays $0.01 per query in HBAR on Hedera — the standard flow of HTTP 402 → pay → 200.
curl -X POST https://<your-host>/resolve \
-H 'content-type: application/json' \
-d '{"task":"get current weather for singapore"}'
# → 402 Payment Required, then settle via x402 and retry → ranked services + who to avoidIt returns the same payThis / avoid / _meta (freshness) shape as the MCP tools.
List your service
A service lists itself by registering an ENS subname of agentindex.eth through our ServiceRegistrar, publishing its endpoint, price and spec as ENS text records.
Listing charges a fee via ServiceRegistrar.setPricing() paid to the registry beneficiary — free during the hackathon (price set to 0).
Names are revocable: a service caught taking payment and returning junk gets delisted on-chain and drops out of the index.