TinyFish

TinyFish

TinyFish 为 AI 代理提供了一种通过单一 API 使用实时网页的方式:搜索、以干净的 markdown 格式获取页面、运行真实浏览器以及完成多步骤任务。所有计划中的搜索和获取操作均不消耗积分。只有当代理需要点击、登录或保持会话时才开始计费。这种划分适用于在动态页面和登录墙上失效的工作。

该类别的大多数网络 API 仅限于搜索和抓取。TinyFish 自身的对比页面将其与 Exa、Firecrawl、Parallel 和 Tavily 并列,然后增加了一个填写表单的网络代理和一个针对阻止抓取器网站的隐身浏览器。计费表保持关闭状态,直到页面需要交互,这种计费方式不同于按抓取次数付费。

DoorDash、Grubhub、The Zebra 和 Google Hotels Japan 的团队使用它进行价格检查、报价和库存管理,这些功能之前没有公开 API。如果你已经运行 Claude、Cursor 或自定义代理堆栈,只需一个 MCP 配置或一个 API 密钥即可。

主要功能:
  1. Web Agent 在 Online-Mind2Web 上的命中率为 89.9%,而 Gemini 2.5 计算机使用为 69%

  2. Fetch 返回 markdown、HTML 或 JSON,有 93% 的测试页面可用

  3. 搜索中位延迟为 556ms,在他们的 1200 查询内部评估中

  4. 浏览器会话冷启动时间不到 250ms,最长持续 60 分钟

  5. MCP 在 agent.tinyfish.ai/mcp 上,支持 Python、TypeScript、CLI 和 REST

  6. Agent 步骤每次消耗 1 点积分;失败的运行不收费

Pros:
  1. 搜索和 Fetch 保持在 0 积分,因此在实时页面上绑定代理不会耗尽积分池

  2. 失败的 Agent 运行收费为 $0,当网站在工作流程中途发生变化时很重要

  3. 一个 API 密钥覆盖搜索、Fetch、浏览器和代理,而无需拼接四个供应商

  4. 已发布的 Mind2Web 和 WebVoyager 分数提供了大多数爬取 API 从未公布的准确性标准

Cons:
  1. 按使用付费将并发 Agent 运行限制为 2 个,浏览器会话限制为 5 个

  2. ISO 27001、SOC 2 和本地部署仅限于企业计划

  3. 浏览器会话限制为 60 分钟,因此长时间门户工作必须分块进行

  4. Agent 和浏览器工作仍然会消耗积分,每步 $0.015 或每 4 分钟浏览消耗 1 积分

常见问题:

Does TinyFish have a free plan?

Yes. TinyFish Search and Fetch are free on every plan and use no credits. New accounts also get 500 free credits for Agent and Browser, with no card required. Paid Starter and Pro plans add monthly credit allotments on top of that.

How much does TinyFish cost?

TinyFish Starter is $15 per month for 1,650 credits, and Pro is $150 per month for 16,500 credits. Pay as you go is $0.015 per credit after 500 free starter credits. Search and Fetch stay at 0 credits. Enterprise pricing is custom.

Does TinyFish have an API?

Yes. TinyFish exposes four APIs: Search, Fetch, Agent, and Browser. Canonical endpoints include GET https://api.search.tinyfish.ai, POST https://api.fetch.tinyfish.ai, and POST https://api.browser.tinyfish.ai. SDKs cover Python and TypeScript, plus CLI and MCP.

What does TinyFish integrate with?

TinyFish connects through MCP, REST, Python and TypeScript SDKs, Claude Code, Cursor, n8n, LangChain, and Dify. The MCP server URL is https://agent.tinyfish.ai/mcp.

Can TinyFish agents log into websites?

Yes. TinyFish Web Agent handles authenticated sessions with persistent browser profiles, so it can return to vendor portals with cookies intact. The Browser API is built for login walls and anti-bot protected sites.

How accurate is the TinyFish Web Agent?

TinyFish reports 89.9% task success on Online-Mind2Web and 91.1% on WebVoyager (641 tasks across 15 live sites). Those are the figures they publish against Gemini 2.5 Computer Use, OpenAI Operator, and BrowserUse.

Is TinyFish a web scraping API?

TinyFish Fetch is a web scraping API that turns a URL into markdown, HTML, or JSON at 0 credits. Search is also free. Agent and Browser are the paid layer for pages that need clicks, forms, or a login.

分类:

定价:

免费试用和收费混合

标签:

Web Scraping
Browser Automation
Stealth Browser
MCP Server
REST API
Data Extraction

使用的技术:

Next.js
Cloudflare
Google Tag Manager
Sanity
Python
Ruby
Typeform
GitHub
Tailwind CSS

评论:

Give your opinion on TinyFish :-

Overall rating

Join thousands of AI enthusiasts in the World of AI!

最佳免费 TinyFish 替代方案(和付费)

By Rishit