Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type @ followed by the MCP server name and your instructions, e.g., "@brainstormform Create a brainstorming form with questions about our product roadmap and open it."
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
Ask a user as many questions as you need, then get out of the way.
Your coding agent can ask you questions, but usually only a handful at a time, one round, no files. Brainstormform is a small local tool that lets an agent open a real form in your browser instead: as many questions as it wants, grouped into sections, with file uploads, conditional follow-ups, and answers that flow back as JSON.
It runs on your machine, in a throwaway local server, with no dependencies and no telemetry.
Built-in "ask the user" tools are built for a quick decision. They fall over the moment a real planning session starts:
The cap is around 4–6 questions, so the agent drops the ones that matter.
There is nowhere to put a section heading or an explanation.
You cannot attach a screenshot, a PDF or a reference image.
It is one round. The agent cannot ask a follow-up based on what you just said.
Brainstormform is what you use when the questions are the point.
Related MCP server: prompt-gui-mcp
Unlimited questions, in sections. A progress bar and pagination keep it navigable.
Every common answer type. Single and multi choice, text, long text, number, rating scale, yes/no, file upload, image cards, a matrix grid and a drag-free ranking question.
Conditional questions. Show a question only when an earlier answer calls for it.
A note on any answer. When none of the options quite fit, add free-text context to any question — even one you skip.
Code and design examples. Attach a snippet or a design mockup to a question, a section or an answer option; open it to see it as code or as a live, sandboxed preview sized for desktop, tablet and phone.
Live follow-ups. The agent reads your answers as you go and can add new questions to the same form while you are still filling it in.
Answers as JSON. Typed and clean, straight into the agent's context. Or ask
for a compact map or Markdown with --format json|md to save tokens.
Personalized questions. A profile you set once from the ⚙ menu — experience level, language, how much detail — is stored locally and shared with every agent, so a senior developer and a new vibecoder get questions written for them.
Local by default. Binds to 127.0.0.1, random port and token, deleted once
read. Nothing leaves the machine.
Requires Node 18.17 or newer.
curl -fsSL https://raw.githubusercontent.com/dutchbase/Brainstormform/main/install.sh | sh
brainstormform setup # wires up Claude Code, Codex and opencode for youOr without installing anything:
npx -y github:dutchbase/Brainstormform ask questions.json --openOr from a checkout: git clone … && cd Brainstormform && npm link. Full details,
including each agent's manual config, are in docs/install.md.
brainstormform ask examples/questions.json --open
# {"sessionId":"bf-...","url":"http://127.0.0.1:PORT/s/TOKEN","pid":1234}
# the form is open in your browser; when you press Finish:
brainstormform wait bf-... --timeout 600Write your own questions with brainstormform guide as a reference, then point
ask at the file. A whole form is just JSON.
This is the part that makes it useful for brainstorming rather than a one-off survey. Answers save as you type, so the agent can keep up:
brainstormform ask questions.json --open
brainstormform progress bf-... # what has been answered so far
brainstormform add bf-... followups.json # add questions based on those answers
brainstormform wait bf-... --timeout 600 # resolves only when you press FinishThe agent can loop progress and add as many times as it wants. New questions
drop into the open form without disturbing what you have already answered. Read
more in docs/live-sessions.md.
Any agent that can run a shell command can use the CLI. For MCP clients there is
also a stdio server, and brainstormform setup detects and configures the ones
it finds:
brainstormform setup # Claude Code, Codex, opencode, Cursor, Gemini CLI
brainstormform setup --target codex # just oneIt installs the skill, writes the MCP entry (backing up any config first), and
runs a health check. Manual snippets for every client live in
docs/mcp.md. The MCP server exposes ask_questions,
read_answers, add_questions and wait_for_answers, and pushes a notification
when you answer.
The skill installed by setup teaches the agent to choose between Brainstormform
and its built-in question tool: small and static goes to the built-in tool, while
more than a few questions, sections, uploads or live follow-ups come here. It
also advises the agent to use the superpowers brainstorming skill as a front
end for creative work — recommended, not required. A copy is bundled in
skills/brainstorming/ and installed by the same command when you don't already
have one. To install them on their own:
brainstormform install-skillsingle · multi · visual · text · textarea · number · scale ·
boolean · file · matrix · rank
Questions support Markdown help text, conditional showIf, and suggestions.
The full format is in docs/schema.md.
No telemetry and no network calls. Nothing leaves the machine.
The server binds 127.0.0.1 only, on a random port, behind a random token.
Foreign Host headers are rejected; upload sizes are capped.
Sessions live in $XDG_RUNTIME_DIR and are deleted the moment the agent reads
them, unless you ask to keep, export or archive them.
Run brainstormform doctor to see a health check and the exact paths the tool
can touch: what it touches on disk.
Releases publish checksums (SHA256SUMS) and are cut from tagged commits.
Live sessions, conditional questions, visual questions, the review screen and
one-command setup have landed. Next up: ranking and matrix questions, a
searchable archive, and a standalone binary so it runs without Node. See
CHANGELOG.md.
Out of scope: hosted or multi-user use. Local-first is the security model.
Contributions are welcome. The codebase is small and dependency-free, so it is a good place to start. See CONTRIBUTING.md.
This server cannot be deployed
Form backend for AI agents: create live form endpoints, configure them, read submissions.
Agent-Native form builder - create, edit, and manage forms
Give AI agents access to form submissions — read, search, update, and process file attachments.
AI-native form builder: create, publish & read responses from Claude, ChatGPT & MCP.
