TinyFish is enterprise infrastructure for AI web agents that automate business workflows across hundreds of sites, reporting a 98.7% success rate.
TinyFish suits enterprise teams that need AI agents to log into sites, fill forms, and pull structured data at scale instead of hand-coding scrapers. It already handles 30 million operations a month for Fortune-scale customers like Google Hotels and DoorDash, replacing bespoke scraping scripts with a managed, natural-language agent runtime.
Parallel execution at 1,000x scale with all-inclusive pricing (no separate browser/proxy/inference costs), Claude MCP integration, and natural language control—enabling non-technical teams to deploy enterprise-grade web automation in minutes, not months.
TinyFish's moat is strong: deep technical execution (82% on Mind2Web vs 43% for Operator), institutional relationships (Google, DoorDash, Fortune 500), and well-capitalized backing ($47M Series A). However, the underlying AI (vision LLMs) is not proprietary, and as frontier models improve at web understanding, competitive moats erode. Lock-in through integrations and workflows provides additional defensibility.
Main trade-off
CAPTCHA handling is unsupported, blocking any workflow on sites relying on image or hCaptcha verification. This is a hard technical limit that affects a significant percentage of modern web properties.
| Plan | Monthly price |
|---|---|
| Free | Free |
| Pay-as-you-go | Free |
| Enterprise | Free |
| Starter | $15/mo |
| Pro | $150/mo |
Free: 500 steps, no credit card required. Pay-as-you-go: $0.015/step, 2 concurrent agents. Starter: $15/month (1,650 steps, 10 concurrent agents, $0.014 overage/step). Pro: $150/month (16,500 steps, 50 concurrent agents, $0.012 overage/step). Enterprise: Custom pricing with custom volume, concurrency, and SLA. All-inclusive: no separate charges for browsers, proxies, or LLM inference.
| Feature | Free | Pay-as-you-go | Enterprise | Starter | Pro |
|---|---|---|---|---|---|
| Web search | 100 req/mo | Unlimited | Unlimited | Unlimited | Unlimited |
| Page fetch (Markdown) | 100 req/mo | Per-use | Unlimited | Unlimited | Unlimited |
| AI browser agents | 5 runs/mo | Per-run | Custom | 100 runs/mo | 1,000 runs/mo |
| Stealth sessions | — | — | ✓ | — | ✓ |
| MCP server | ✓ | ✓ | ✓ | ✓ | ✓ |
TinyFish is enterprise infrastructure that lets organizations run autonomous AI agents across external websites: logging in, filling forms, and extracting structured data without managing browsers or proxies. It scores 82 percent on the Mind2Web benchmark, well ahead of OpenAI's Operator, and already powers production traffic for Google Hotels and DoorDash.
Parallel Web Automation
Run up to 1,000 simultaneous operations across multiple websites in a single deployment, enabling massive-scale web data operations without managing infrastructure.
Natural Language Control
Describe web tasks in plain English: no CSS selectors, XPath, or code required. Agents interpret goals and execute workflows autonomously.
Authentication and Dynamic Sites
Navigate password-protected sites, handle multi-factor authentication, interact with form-based systems, and extract data from JavaScript-rendered content at production speed.
Claude MCP Integration
Connect TinyFish directly to Claude Desktop via Model Context Protocol, enabling AI assistants to run web automation with real-time progress streaming.
AgentQL Query Language
Extract structured data and automate workflows using AgentQL, a specialized query language with Python and JavaScript SDKs, REST API, and browser debugger.
All-Inclusive Pricing
No hidden charges for browsers, proxies, residential IP routing, or LLM inference. Step-based billing with transparent overage costs declining by tier.
Most users expect these and they are not currently supported.
| Tool | Price from | Free plan | Runs on | Rating |
|---|---|---|---|---|
| $15/mo | Yes | Web | 3.7 | |
| $49/mo | Yes | Web | 4.0 | |
| $39/mo | Yes | Web · Mac · Windows | -- | |
| $17/mo | Yes | iOS · Android · Mac · Windows | 4.0 |
Proprietary
TinyFish charges by step: pay-as-you-go runs $0.015 per step with 2 concurrent agents, Starter is $15 a month for 1,650 steps and 10 concurrent agents ($0.014 overage), and Pro is $150 a month for 16,500 steps and 50 concurrent agents ($0.012 overage). Enterprise pricing is custom, and every plan bundles browser, proxy, and AI-inference costs into the same per-step rate.
The free tier includes 500 steps a month with no credit card required, covering the same authentication, extraction, and MCP access as the paid plans. Once a workflow crosses 500 steps, TinyFish switches automatically to pay-as-you-go billing or a paid tier.
Airtop is a simpler pick for no-code, single-site browser automation, and Hexomatic or Axiom AI cost less if you only need to automate one site at a time. TinyFish is built for enterprise-scale parallel automation across hundreds of sites with native Claude integration, so it fits teams that have outgrown single-site tools.
Airtop is the simpler, no-code option and works well for individual sites, but it does not offer TinyFish's parallel execution across many sites at once or the all-inclusive step-based pricing that folds browser and proxy costs into one rate. TinyFish is the stronger choice once a team needs enterprise-scale automation rather than a single workflow.
Sign up for the free tier (no credit card needed) and either call the REST API directly or connect through the Python or JavaScript SDK. TinyFish also ships an MCP server, so Claude Desktop users can connect and start running agents in a chat interface within minutes.
TinyFish is enterprise web agent infrastructure that launched from stealth in August 2025 with $47 million in Series A funding. It was founded in 2023 by Sudheesh Nair (former Nutanix President), Shuhao Zhang (ex-Meta engineering leader), and Keith Zhai (ex-Wall Street Journal), and enables organizations to automate complex web interactions at scale without managing browsers, proxies, or infrastructure themselves.
The platform runs as a serverless API: it accepts a natural-language goal and a target URL, then dispatches an AI agent to navigate the site, handle logins and multi-step forms, extract structured data, and complete the transaction while streaming progress in real time. Traditional web scraping tools need CSS selectors, XPath, or custom configuration; TinyFish agents work from a plain-English task description instead, connecting to Claude Desktop through the Model Context Protocol and exposing a REST API, Python and JavaScript SDKs, and an AgentQL query language for developers who want more control.
On the Mind2Web benchmark, TinyFish scores 82 percent versus 43 percent for OpenAI's Operator, a 39-point lead the company points to as evidence of production readiness rather than a research demo.
TinyFish already runs in production at large enterprises: Google uses it to aggregate hotel inventory across Japan for Google Hotels search, DoorDash relies on it to manage web-interaction complexity, and ClassPass, Amazon, Amplemarket, NextEra, Jobright, and Healdev deploy it across healthcare prior-authorization tracking, insurance regulatory-filing monitoring, financial services, retail price monitoring, and supply-chain shipment tracking.
Product facts on this page (pricing, plans, integrations, compliance) are read from TinyFish's own pages and documentation. The verdict is Hokai's reading of them, not the vendor's claim.