Taste is a pay-per-call MCP server that puts vetted human experts in the loop for AI agents.
What it does — humans review an AI agent's output across any media (text, images, video, audio, social posts), audit step-by-step reasoning chains for mid-chain errors, run prepublish safety checks (cultural sensitivity, brand safety, derivative risk), gate high-stakes actions for explicit human approval before anything irreversible happens, and convene curated expert teams for 30- or 60-minute think-tank working sessions on an idea, problem, or decision. Approved outputs receive a Taste content certificate recorded on Base — any downstream agent can verify the review on-chain.
Connect — MCP endpoint https://api.humantaste.app/mcp/ over Streamable HTTP (MCP 2025-06-18). No API key, no signup. Free tools (list_offerings, list_expert_profiles, get_result, verify_certificate) answer immediately. Paid tools cost $1.00–$15.00 per call (evaluation tools $1, think tank $5/$15), settled in USDC on Base mainnet via x402 — call without payment, get an HTTP 402 challenge, sign an EIP-3009 USDC transfer, retry. Typical human turnaround is 5–30 minutes for evaluation, up to 24h for a think tank session.
Paid tools — verify_external_source, review_content, verify_reasoning_chain, prepublish_review, predict_audience_reaction, arbitrate_dispute, request_human_approval, order_think_tank_session_30, order_think_tank_session_60.
Compliance — EU AI Act Article 50(2) exempts AI-generated content from mandatory AI-disclosure labels when it has had human review or editorial control. Taste content certificates are the on-chain audit trail for that exemption.
Links — one-pager https://api.humantaste.app/llms.txt · full reference https://api.humantaste.app/llms-full.txt · brand https://humantaste.app
服务器配置
{
"mcpServers": {
"taste": {
"url": "https://api.humantaste.app/mcp/"
}
}
}