Many simulated users.
One picture of your app.
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.
To join the waitlist, email contact@hokusai.ai.
No terminal. No config file. No install. If you can paste a link, you can run it.
You bring
Your app’s link.
Nothing to configure.
Paste your app’s link, MCP or website. Pick a test, and Hokusai works out the rest.
You skip
- Config files
- Test scripts
- Writing test cases
- A terminal
- Installing anything
- Code on your server
How it works
Step 1 · Add an app
Paste a link. That’s the setup.
Give your app a name and paste its link. Hokusai connects, checks it can reach your app, and finds every tool it has on its own.
Paste a link. Hokusai finds every tool on its own.
- Nothing to install on your server, no code to add.
- Needs an access key? Paste it too. That’s the only extra.
- Unreachable? It says so plainly, and says what it tried.
Step 2 · Choose a run
Pick a size. Name what matters.
Every profile comes ready to run, from a three-agent Check to a 200-agent Initial QA. The only thing you write is the journeys you care about, one line each, in plain words.
Ready-made sizes. Name your journeys, see the price first.
- No settings to tune. Change the mix only if you want to.
- Hokusai explains who will test the app and how.
- The price is on the Start button before you press it.
Step 3 · Run
Watch it happen, live.
Agents launch in waves. See what each one is doing right now, how far along the run is, and what it has cost so far.
Watch every agent, the cost, and issues as they appear.
- Issues appear as soon as they are found.
- Stop at any time. Whatever was gathered is kept and reported.
Step 4 · Read
One verdict. All the evidence.
A verdict, scores from 1 to 5, issues ranked worst first, what worked well, and suggestions. Every finding links to the moment that supports it.
A verdict, scores, ranked issues, and the evidence.
- Cut-short runs say so in the verdict.
- Every run ends with a four-line receipt.
Scroll to play · scroll back to rewind
Add an app
Streamable HTTP. Hokusai connects and lists the tools when you add it.
Sent as a bearer token on every call to your server.
One or two sentences help the Real User agents pick realistic goals.
Add an app
5 tools discovered
- Input schema
search_hotelsFind hotels by city, dates and price. - Input schema
get_hotelDetails, rooms and prices for one hotel. - Input schema
start_bookingHold a room while the guest decides. - Input schema
create_bookingConfirm a held room and return a payment link. - Input schema
cancel_bookingCancel a booking and release the room.
New run
1App
https://travel.example/mcp·5 tools·Connected2Profile
A starting point for the mix below. Adjust any count afterwards and the run becomes custom.
3Agents
Who tests your app, and how many of each. Agents of the same type share a brief but act independently.
Methodical testers. Map the whole app from your tools, then check each use case: happy path, bad input, edge cases, what the errors say.
The journeys you name, in plain words. Each one gets its own QA tester, or several working independently, who try to get through it end to end.
+ 17 more journeys
People with a goal and a persona, no manual. Try to get something done the way a person would through an assistant: conversationally, impatiently, without reading descriptions first.
Chaos. Odd orders, malformed and extreme inputs, repeats, abandoned flows. Finds crashes and inconsistent state.
4Concurrency
How many agents run at the same time. Lower is gentler on your server. Higher finishes sooner.
80 agents in 5 waves · about 12–16 min
5About your app (optional)
One or two sentences help the Real User agents pick realistic goals. Leave it empty and they will work it out from your tools.
Run summary
- Agents
- 80
- At a time
- 16
- Estimated time
- 12–16 min
- Tokens est.
- $36.54
- Run fee
- $5.00
- Agent fee
- $52.50
- Concurrency
- $7.50
- Total
- $101.54
- Balance
- $120.00
- After this run
- $18.46
Agents call your server's real tools. Don't point a Monkey Test at production data.
Start run · $101.54Travel assistant MCP · Run #27
All 80 agents finished
Reviewers are reading the sessions and writing the report
- 14:08
- elapsed
- $96.40
- cost so far
- 2,184
- tool calls
- 11
- issues so far
Agents
live| Agent | Status | Now |
|---|---|---|
| QA #3Use cases from the tool list | Done | Finished |
| Journey #3Change the dates of a booking | Done | Finished |
| Real User #12Ana, in a hurry | Gave up | Gave up after 2 tries |
| Real User #23Tomás, comparing prices | Done | Finished |
| Monkey #4Abandoned flows | Done | Finished |
Findings so far
unverified until reviewedTravel assistant MCP · Run #27
Overall score across 4 dimensions
Search and comparison worked for almost everyone. Booking is the problem: create_booking rejected valid payloads in 23 of 25 Real User sessions, and abandoned bookings kept rooms held.
Journeys
17 of 20 completedCancelled on the first try; the room went back on sale straight away.
create_booking rejected the hotel_id that search_hotels had just returned.
Evidence: Journey #1 · step 6
Issues
11 found · worst firstAgents sent a hotel_id taken from search_hotels. Expected: a confirmation.
create_bookingfound by 23 of 25 Real User6 of 20 Journey QAA booking started and never finished kept the room off sale. Expected: the hold to expire.
start_bookingfound by 9 of 25 Monkey TestScores
1 to 5Receipt
Settled- Tokens
- $35.90
- Run fee
- $5.00
- Agent fee
- $52.50
- Concurrency
- $7.50
- Charged
- $100.90
- Reserved at start
- $101.54
Reports
Every issue comes with proof.
The proof is the call.
Highcreate_booking rejects valid payloads
→ create_booking {"hotel_id": "h_2231"}← "invalid hotel_id" 388 ms
“That id came straight from search_hotels. Trying once more.”
The proof is the screen.
LowBack and Pay buttons sit out of line
Deposit
Your table
“The two buttons don't line up. Back looks like it slipped.”
step #4 · 1440 × 900
The proof is the widget.
Mediumhotel_card shows a broken photo
Find me a hotel in Lisbon for Friday
Used search_hotels
Casa Alfama is the best match near the river:
“The photo never loaded, just a broken icon. I'd want to see the room before I book.”
chat 380 px · Real User #9
- What went right, not just problems
- Honest verdicts when a run is cut short
- Every run kept, to compare and share
Run specs
Five kinds of tester.
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.
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.
Coming to the agent library later:UI/UXAccessibilityContentPerformanceLocalisation
Run sizes
From a quick check to a crowd.
Check
A tester on one journey, a real user and a monkey. For every pull request.
$5.65 per runJourney QA
The journeys you name. For a feature or a flow you are unsure about.
$6.36 3 journeysGeneral QA
The whole app, mapped from scratch. For a new app nobody has mapped.
$18.74 per runFull QA
General, 15 to 20 journeys, Real User and Monkey Test. The weekly pass.
$101.54 per runInitial QA
Full QA with more personas, more journeys, a longer Monkey Test. Before a first release.
$223.30 200 agentsThe price the Start button shows for each run: estimated tokens plus fees. A run is settled for the tokens it actually used.
Why not just hire testers?
A tester's day buys one pass. Hokusai buys a crowd.
Your first hundred users, before your first real one.
Pricing
Pay for runs. Not seats.
No plan, no subscription. A prepaid balance, and every run priced before it starts.
- Check$5.65
- General QA$18.74
- Full QA$101.54
See your app through fresh eyes.
Nothing to configure. Paste a link, get a report.Hokusai is not public yet. Join the waitlist and we will email you when it opens. New accounts start with $5 of runs.
To join the waitlist, email contact@hokusai.ai.
No terminal. No config file. No install.