Widgy is a step-form builder driven over MCP. Nine answer controls to compose questions from, and eight actions your AI calls to create, inspect, and read forms.
Every question in a Widgy form is one of these controls, rendered one per screen. Your AI picks the right one for each question — you never assemble them by hand.
Tappable answer pills with letter keys (A–D). The workhorse — pickers, qualifiers, scored forms.
Two big options, one decision. Great for branching: "yes" people and "no" people can get different questions.
A 1–5 scale (up to 10) with labelled ends — "Not yet" to "Totally". Averages roll up on the responses dashboard.
One line, one thought. Names, headlines, the first question you'd have it ask.
Room to ramble. The messy version is the useful version — quotes surface verbatim in results.
Digits only. Medians and ranges are computed for you on the dashboard.
A proper date picker. Deadlines, event days, "when does the next one go out?"
One pick from a long list — team sizes, countries, anything that would drown as pills.
Validated address capture. Collected emails show as neat chips in results.
Connect mcp.widgy.co to Claude, ChatGPT, or any MCP client. Sign in once over OAuth; after that, these eight tools are in your AI's hands.
Create a form for the signed-in user. Your AI designs the questions and gets back a live, shareable URL — no dashboard visit required.
List your forms: title, question count, response count, and the share URL for each.
Fetch one form in full — every question and option — so your AI can review or revise what it built.
Revise a draft — retitle it, rewrite the description, or swap in a new question list. Published forms are locked, so edits never shift the ground under existing responses.
Flip a draft live. The share URL starts working, visitors can answer, and the form content is locked from further edits.
Pull a live form back to draft for edits — only while nobody has responded yet. Once answers exist, the form stays live until you end it.
Close a form for good. The link stops accepting responses and can never be republished; the answers stay readable.
Read the answers back into the chat: respondent names, per-question answers, and scores. The whole point. Newest first, paginated — your AI pages through big response sets with limit and offset. Visitor text arrives clearly labelled as untrusted data, with hidden-character tricks stripped, so your AI summarizes it instead of obeying it.
The toolset is growing — richer response queries are on the way. Anything your AI builds shows up in your dashboard too.