TinyFish

TinyFish is an AI-powered web agent platform built for teams that need live browser actions, not just static data scraping. Instead of writing brittle selectors or XPath, you describe a goal in plain English—like “log in, search for this product, and extract the price”—and TinyFish navigates real websites, handles authentication, and completes multi-step workflows. It’s delivered through a visual workbench, an API, and MCP (Model Context Protocol) integration, with a strong focus on reliability at scale. Designed for developers, operations teams, and enterprise workflows, TinyFish is infrastructure for browser automation, not a one-off experiment.

• Natural language control: Give instructions in plain English; TinyFish translates them into browser actions without manual selectors or JavaScript.
• Live execution streaming: Watch runs in real time with streaming events and a live browser preview, making debugging straightforward.
• Stealth and proxy options: Anti-detection browser modes and geo-targeted residential proxy support help access harder-to-reach or protected sites.
• Parallel agent runs: Run from 2 up to 50 concurrent agents, allowing high-volume, simultaneous web operations.
• Workbench, API, and MCP access: Build workflows visually in the Workbench, integrate via REST API, or connect through MCP for assistant-driven automations.
• Built‑in anti‑bot and observability: Screenshots, run history, and anti‑bot protection are included, so you can track what happened and why.
• Bundled infrastructure costs: Browser, proxy, LLM, and anti‑bot costs are wrapped into the per‑step pricing—no separate surprise bills.

1. Sign up and access the environment: Create an account at tinyfish.ai and open the visual Workbench or get your API keys for programmatic access.
2. Describe your task in plain English: Start a new run and write a natural‑language instruction (e.g., “Go to acme‑shop.com/login, use these credentials, search for SKU 12345, and return the price”).
3. Configure run settings: Choose the number of parallel agents, select anti‑detection browser mode if needed, and set proxy locations for geo‑targeting.
4. Execute and monitor live: Launch the agent and watch the real‑time browser preview and streaming steps; you’ll see exactly where the agent clicks and types.
5. Review outputs and history: After completion, inspect screenshots, extracted data, and detailed logs in the run history view. Debug any issues by replaying steps.
6. Integrate into your stack: Use the REST API to trigger runs asynchronously or connect TinyFish to your AI assistant through MCP for agent‑driven orchestration.

TinyFish uses a per‑step model where every action (click, type, navigation) counts as one step, and all supporting costs—browser, proxy, LLM inference, and anti‑bot protection—are bundled into the price. Below is an overview of the available tiers. Please note that pricing details may not be up to date; always check the official TinyFish website for current information.

• Pay‑as‑you‑go: $0.015/step. Includes 2 concurrent agents, remote browser, residential proxy, LLM costs, anti‑bot protection, and 30‑day run history. Email support.
• Starter: $15/month. 1,650 steps/month, 10 concurrent agents, all bundled costs, 30‑day history with observability & screenshots, priority email support, Workbench/API/MCP access. Overage $0.014/step.
• Pro: $150/month. 16,500 steps/month, 50 concurrent agents, 180‑day history, priority email + Slack/Discord community, Workbench/API/MCP. Overage $0.012/step.
• Enterprise: Custom pricing. Tailored step volumes, concurrency, advanced security/compliance, dedicated support engineer, custom SLA, and on‑premise deployment options.

• Web Workbench: A visual, no‑code interface where you can design, run, and monitor agents without writing code.
• REST API: Full programmatic access for triggering runs, fetching results, and integrating TinyFish into your own applications or pipelines.
• MCP (Model Context Protocol) integration: Connect TinyFish directly to AI assistants and agent frameworks so they can execute live web actions as part of broader workflows.
• Remote browser engine: Browser sessions run on TinyFish’s infrastructure, so there’s no local setup or maintenance overhead.

• Use anti‑detection modes proactively for sites behind bot‑blocking services; combine with residential proxies when geo‑specific access matters.
• Keep natural‑language instructions precise: Mention the exact elements to interact with (e.g., “click the blue ‘Sign In’ button”) to reduce ambiguity.
• Start with a single agent to verify the flow works, then scale up to concurrent agents for production‑level volume.
• Leverage the live preview during initial runs—it’s the fastest way to spot mis‑clicks or login failures and refine your instructions.
• Plan around CAPTCHA limitations: TinyFish cannot automatically solve CAPTCHAs, so avoid relying on workflows that hit them frequently or use sessions where login bypasses them.
• Use the bundled cost model to your advantage: Because LLM, proxy, and browser costs are included per step, focus on optimizing the number of steps rather than worrying about separate metered bills.

• Developers and automation engineers who need a reliable API‑first web agent that can handle authentication and complex multi‑step tasks.
• Ops and SRE teams wanting to automate checks against live internal dashboards or third‑party portals without building custom Puppeteer scripts.
• E‑commerce and retail operations tracking pricing, checkout, and inventory across many sites concurrently.
• Data extraction and growth teams who need structured data from messy, real‑world websites and prefer describing the task in English over maintaining scrapers.
• Enterprise architects looking for an on‑premise‑capable, auditable automation layer that integrates with existing AI and workflow tools.
• Technical product managers evaluating AI web agents for production use, not just prototyping, thanks to the observability, scaling, and security features.