# Hokusai > Hokusai is an AI QA team on demand. Dozens of simulated users work through your app at once, and you get a scored report on what holds up and what doesn’t. No terminal. No config file. No install. If you can paste a link, you can run it. Hokusai is not public yet; join the waitlist at https://hokusai.ai/#join. ## What it is Paste your app’s link, MCP or website. Pick a test, and Hokusai works out the rest. - Nothing to install on your server, no code to add. Needs an access key? Paste it too. That’s the only extra. - What you skip: config files, test scripts, writing test cases, a terminal, installing anything, code on your server. - Each tester already knows what to try, so there are no test cases to write. Pick a profile, adjust the mix, or write your own. - Unreachable? It says so plainly, and says what it tried. - Agents launch in waves. You see what each one is doing, how far along the run is, what it has cost so far, and issues as soon as they are found. Stop at any time; whatever was gathered is kept and reported. ## What it tests - **MCP servers**: paste the server’s link. Hokusai connects, checks it can reach it, and finds every tool it has on its own. The proof is the call: the request, the response, and what the agent was thinking. - **Websites**: paste the site’s link. The proof is the screen: the tester’s screenshot of the step. - **MCP widgets**: a server’s UI widget as it renders inside an AI chat. The proof is the widget, as the tester saw it in the chat. ## Before you submit Directories and stores check the same things every time. Hokusai runs that list against your app and tells you what would come back. - **Claude connector directory** (MCP servers): Every tool titled and annotated read-only or destructive. Descriptions that match what the tool does. A privacy policy, and only the data it needs. - **ChatGPT Apps Directory** (MCP servers): Annotations ChatGPT trusts when it decides to ask before acting. Descriptions that don’t overpromise. The safety, privacy and transparency rules of the submission guidelines. - **App Store** (Mobile apps, later): Crashes, dead ends and flows that don’t finish, found the way a reviewer finds them in the first minutes. - **Google Play** (Mobile apps, later): Permissions the app doesn’t need, broken flows, and policy text that is missing. ## Run specs - **General QA**: Careful testers who map the whole app. What they do: Work out your use cases from the tools, then try each one, edge cases included. What gets graded: Whether it works, how it handles errors, how clear the tools are. - **Journey QA**: The same testers, with your list. What they do: Run each journey you name, like “sign up and log in”, end to end. What gets graded: Whether each journey finished, and where it broke. - **Real User**: A goal, a persona, no manual. What they do: Try to get something done through an assistant, impatiently, without reading the docs. What gets graded: Whether they got what they came for, and would trust it. - **Monkey Test**: Chaos, on purpose. What they do: Call tools in odd orders with broken inputs, and quit flows halfway. What gets graded: Crashes, hangs, raw errors and broken state. - **Custom**: Your tester, your brief. What they do: Follow a brief you write, for anything the other four miss. What gets graded: The criteria in your brief, with the same evidence. Coming to the agent library later: UI/UX, Accessibility, Content, Performance, Localisation. ## Run sizes Every profile comes ready to run. The only thing you write is the journeys you care about, one line each, in plain words. Prices as of 2026-11-03. - **Check**: 3 agents, 2 at a time; $5.65 per run. A tester on one journey, a real user and a monkey. For every pull request. - **Journey QA**: 1 agent per journey, 4 at a time; $6.36 for 3 journeys. The journeys you name. For a feature or a flow you are unsure about. - **General QA**: 10 agents, 4 at a time; $18.74 per run. The whole app, mapped from scratch. For a new app nobody has mapped. - **Full QA**: 80 agents, 16 at a time; $101.54 per run. General, 15 to 20 journeys, Real User and Monkey Test. The weekly pass. - **Initial QA**: 200 agents, 32 at a time; $223.30 per run. Full QA with more personas, more journeys, a longer Monkey Test. Before a first release. Up to 300 agents in one run. The price is on the Start button before you press it: estimated tokens plus fees. A run is settled for the tokens it actually used. ## Pricing - Pay for runs, not seats. No plan, no subscription: a prepaid balance, and every run priced before it starts. - A run’s price is model tokens for the agents and reviewers, at cost, plus three fees: a run fee, a fee per agent, and a fee per agent running at once beyond the first. - New accounts start with $5 of runs. - The pricing calculator prices any run and sets a year of runs against the cost of hiring one tester: https://hokusai.ai/calculator/ ## What a report contains - A verdict and scores from 1 to 5. - Issues ranked worst first, what went right (not just problems), and suggestions. - Every finding linked to its evidence: the call, the screen, the widget. - An honest verdict when a run is cut short. - Every run ends with a four-line receipt. - Every run kept, to compare and share. ## Compared with a human tester | Aspect | One tester | Hokusai | | --- | --- | --- | | Setup | Weeks of onboarding | Paste a link (nothing to configure or install) | | People | 1 tester | Up to 300 at once (every kind of tester in one run) | | Time | 2 to 3 days | About 15 minutes (80-agent Full QA, start to report) | | Cost | A day's pay | $101.54 for 80 agents (or $5.65 for a Check) | | Availability | Office hours | Whenever you press start (3 a.m., Sundays) | Your first hundred users, before your first real one. ## Why Hokusai - **We measure agents before we trust them.** Every tester and every reviewer is run against real apps and scored before it ships. A spec that cannot show its evidence does not go in. - **We have done this for frontier models.** The team has worked on frontier models with the labs that build them. Hokusai is that work, turned toward your app. - **One honest picture, not a pile of transcripts.** Many agents and many reviewers look at the same app, and the result says what it does not know. ## Status Hokusai is not public yet. Join the waitlist at https://hokusai.ai/#join and we will email you when it opens. ## Links - [Hokusai](https://hokusai.ai/): the product, a run from start to report, the submission check, run specs, run sizes, why Hokusai, and pricing. - [Pricing calculator](https://hokusai.ai/calculator/): build a run, see its price, and compare a year of runs with hiring one tester.