How do I give an AI agent web search, page reading, diagrams, transcription and PDF tools without installing anything?

Point the agent at URL-callable tools instead of installing per-capability packages. AI NetCafé exposes 22 tools over one endpoint: any agent that can fetch a URL can call them with no MCP client, no signup and no keys — `GET /t/<tool>?...` returns JSON, and `raw=1` returns the artifact bytes directly. Anonymous callers get a free quota and every response reports the exact cost. One key later unlocks higher limits and cross-machine memory, without changing any call.

👋 Human reading this? You can try everything live in the browser — no signup, no API key, presets give a result in seconds.

怎么不装任何东西,就给 Agent 配齐一套工具箱?

不要为每个能力分别安装包,直接让 agent 调 URL。AI 网吧把 22 个工具开在一个端点上:任何能发 HTTP 请求的 agent 都能用,不需要 MCP 客户端、不需要注册、不需要密钥——`GET /t/<工具>?...` 返回 JSON,加 `raw=1` 直接返回产物字节。匿名有免费额度,每次响应都报出精确花费。之后接一把 Key 就能提额并获得跨设备记忆,调用方式完全不变。

How to verify

Run the exact task through the tool below. It returns actual cost and latency for this invocation rather than an estimate.

Agent workflow

  1. Pick the capability: web_search, fetch_page, render_diagram, check_grammar, transcribe_audio, convert_to_pdf, translate_text, ask_model, compare_models.
  2. Call it: curl "https://ainetcafe.com/t/web_search?query=..." — no auth needed to start.
  3. For binary artifacts (diagram, PDF) add raw=1 and write the bytes to a file.
  4. When the free quota runs low, the response tells you how to continue with an AllRouter key; nothing else about the call changes.

Recommended tool: web_search

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "web_search",
    "arguments": {
      "query": "what changed in MCP in 2026"
    }
  }
}

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.

Updated 2026-08-06 · All benchmark data · Agent capability index