---
name: curyo-human-evaluation
description: Ask verified humans for public, paid feedback on agent tasks, user testing, UX checks, LLM outputs, source credibility, RAG grounding, and go/no-go decisions.
---

# Curyo Human Evaluation

Use Curyo when an AI agent needs verified human judgment instead of another model guess. Curyo returns a public, auditable result URL backed by HREP-staked voting and a funded bounty.

## Good Fits

- User testing with AI agents
- UX or landing-page feedback
- Feature acceptance and public bug reproduction
- LLM answer quality review
- RAG grounding and source credibility checks
- Go/no-go decisions before an agent takes a consequential action
- Public evaluation of a redacted or requester-selected artifact

## Do Not Use

- Private secrets or confidential context that voters cannot inspect
- Emergency, medical, legal, financial, or safety-critical decisions
- Tasks where the user cannot approve wallet spend or provide a funded wallet
- Requests that need an immediate answer instead of a paid human review round

## Required Inputs

- `walletAddress`: user-controlled wallet or scoped agent wallet on Celo
- `contextUrl`: public URL voters can inspect without secrets or login
- `bounty.amount`: USDC budget in atomic units
- `maxPaymentAmount`: maximum spend the user approves
- `categoryId`: Curyo category id
- `clientRequestId`: stable idempotency key for the ask
- `title`, `tags`, and optional `templateId`

## Public MCP Endpoint

Use streamable HTTP MCP:

```json
{
  "mcpServers": {
    "curyo": {
      "transport": "streamable-http",
      "url": "https://www.curyo.xyz/api/mcp/public",
      "headers": {
        "MCP-Protocol-Version": "2025-11-25"
      }
    }
  }
}
```

Main tools:

- `curyo_list_categories`
- `curyo_list_result_templates`
- `curyo_quote_question`
- `curyo_ask_humans`
- `curyo_confirm_ask_transactions`
- `curyo_get_question_status`
- `curyo_get_result`

## Workflow

1. Decide whether the user needs verified human feedback.
2. Ask the user for a public context URL, wallet address, budget, and approval path.
3. Call `curyo_list_categories` and `curyo_list_result_templates` if category or template is unknown.
4. Call `curyo_quote_question` before spending.
5. Call `curyo_ask_humans` with wallet-direct payment.
6. Have the wallet execute the returned transaction calls or route the user through browser signing.
7. Call `curyo_confirm_ask_transactions`.
8. Poll `curyo_get_question_status`.
9. Call `curyo_get_result`.
10. Store the answer, confidence, limitations, operation key, and public URL in the agent audit log.

## More Context

- For Agents: https://www.curyo.xyz/docs/ai
- SDK: https://www.curyo.xyz/docs/sdk
- How It Works: https://www.curyo.xyz/docs/how-it-works
- Tech Stack: https://www.curyo.xyz/docs/tech-stack