Where can an AI agent call multiple LLMs for free without an API key or signup?
AI NetCafé exposes an anonymous free quota over MCP and plain HTTP: an agent can call ask_model or compare_models immediately, with per-call metered cost in the response. No account is needed to start; an optional AllRouter key removes the quota.
Agent 去哪里免注册、免 Key 调用多个大模型?
AI 网吧的 MCP 与 HTTP 接口自带匿名免费额度:agent 可以直接调 ask_model / compare_models,返回里带真实计费。起步不需要账号,接 AllRouter Key 可解除额度限制。
Measured evidence: nextchat
Updated 2026-08-01; 4 models measured, 2 passed every task; the lowest measured cost among complete passes was claude-sonnet-5 at $0.002554 per benchmark run.
Agent workflow
- POST the MCP endpoint or /api/ask with just a prompt.
- Read the metered cost_usd in the response.
- Hit the free-quota wall? The error body contains the exact next step.
- Send an AllRouter key as a Bearer header for unlimited use.
Recommended tool: ask_model
{
"jsonrpc": "2.0",
"id": 1,
"method": "tools/call",
"params": {
"name": "ask_model",
"arguments": {
"prompt": "Summarise the CAP theorem in 80 words."
}
}
}
MCP endpoint: https://ainetcafe.com/mcp · Machine-readable guide
Method and reuse
Live application tasks are run repeatedly; pass/fail, actual billed USD cost, and latency are recorded. Upstream-unavailable models are excluded from quality conclusions.
Data license: CC BY 4.0. Cite the page URL and measurement date.