Testable MCP: No interface to learn. Just ask ChatGPT, Claude or Grok.
Describe your study in the chat. It lands in your Testable account, ready to pilot.
Connect via MCPMCP is how your AI app signs in to Testable. New here? Create a free account
| rowNo | stim1 | color | keyboard |
|---|---|---|---|
| 5 | BLUE | stim1:green | bg |
| 7 | GREEN | stim1:blue | bg |
Your study ran overnight. Ask what it found. A class-ready task. Before the lecture starts. Any study for your project or dissertation. Built from the chat.
Give me mean RT and accuracy by condition.
Also ask Add a practice block with feedback to my Stroop task.
| Condition | Mean RT | Accuracy |
|---|---|---|
| Congruent | 612 ms | 97% |
| Incongruent | 698 ms | 91% |
Shorten my Stroop task to 20 trials for the methods class.
Also ask Summarize the class results by condition.
| # | stim1 | color | keyboard |
|---|---|---|---|
| 2 | BLUE | stim1:blue | b |
| 3 | GREEN | stim1:blue | b |
| 4 | GREEN | stim1:green | g |
| 5 | BLUE | stim1:green | g |
Build my dissertation study: a lexical decision task with a consent form.
Also ask Add a short demographics survey at the end.
| Part | Size |
|---|---|
| Consent form | 1 page |
| Lexical decision | 160 trials |
| Demographics | 4 questions |
The result is a project in your Testable account, just like one you build by hand. Open it and pilot it yourself.
Any paradigm. Not a template.
When no ready-made task fits, your AI app writes one with the Testable JavaScript SDK. Mouse tracking, drawing, games. The data land in your results like any other trial.
How HTML tasks workSaved in your results
- response
- blue
- rt
- 742
- path
- 86 points
mouse-stroop.html, written by your AI app
It works as you. Your studies stay protected.
3 identifiers never sharedIP address, session id, completion code.
1 confirmation from youbefore a live study with results changes.
0 changes to Testable MindsRead-only: it sees your studies and budget.
For your ethics board
The researcher's own AI app (ChatGPT, Claude or Grok) works under their Testable account and sees only their projects. Participant IP addresses, session ids and completion codes are never passed to it. Trial responses the researcher asks about are processed by that app under its terms.
Quick answers.
What is Testable MCP?
The link between your AI app and your Testable account. MCP (Model Context Protocol) is the open standard that lets an AI app work inside another service. Testable Magic builds inside Testable. Testable MCP works from the AI app you already use.
Can ChatGPT, Claude or Grok build my experiment?
Yes, and you can check it. The experiment is a project in your Testable account, just like one you build by hand. Open it, pilot it in your browser, and change anything by hand or in the chat.
Does my AI app see participant data?
It reads the results of your own projects to answer you, under that app’s terms. IP addresses, session ids and completion codes are never passed to it.
What does it cost?
Nothing extra. Testable MCP is part of every Testable account, with no waitlist. Your AI app may need a plan that allows custom connectors.
Sign in. Click Connect. Done.
No code, nothing to install, nothing to learn. Add Testable in your AI app and sign in.
Connect via MCP See the steps for your app