MCP Questionnaire
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., "@MCP QuestionnaireCreate a prefilled onboarding form for the new hire, with assumptions listed."
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.
MCP Questionnaire
Structured input surfaces inside the Claude conversation — prefilled forms, assumption ledgers, review matrices — instead of clarifying questions as prose bullets. The agent infers what it can, renders a form already prefilled with provenance on every value, and your job is auditing a proposal, not authoring one. Answers flow straight back into model context.
DESIGN.html is the canonical spec (open it in a browser — the mockups are clickable). This README only tells you how to install and run.
Install (Claude Code)
One install gives you the connector and the skill that knows when a form is the right move:
/plugin marketplace add JakubMMazurek/MCP-Questionnaire
/plugin install mcp-questionnaire@mcp-questionnaireor from a shell:
claude plugin marketplace add JakubMMazurek/MCP-Questionnaire
claude plugin install mcp-questionnaire@mcp-questionnaireRelated MCP server: Thinking Patterns MCP Server
Install (claude.ai / Claude Desktop — where the forms render)
Settings → Connectors → Add custom connector → paste the URL from
packages/plugin/.mcp.json. Claude will send you
to GitHub to sign in; if your GitHub login is on the server's allowlist you come
back connected. First render asks a one-time "Always allow" permission. Add on
web/desktop first — mobile picks it up from there.
Access control
The connector URL is public and carries nothing. Every MCP request needs an
OAuth token, and the only way to get one is to sign in with a GitHub account
whose login appears in GITHUB_ALLOWED_USERS in
packages/worker/wrangler.jsonc —
comma-separated, matched case-insensitively. Anyone else gets a 403 and no
token.
Add a teammate: add their GitHub login to GITHUB_ALLOWED_USERS and
redeploy the Worker. Remove one: delete the entry and redeploy; that blocks
new authorizations at once, and any token they still hold expires on its own (or
delete their grant:/token: keys from the OAUTH_KV namespace to cut it
short).
This is a deliberate reversal of the original "no auth" decision (DESIGN.html §3): the URL path used to be the capability, which only worked while this repo stayed private, and that stopped being true once the connector had to be installed from a plugin marketplace. Form ids remain capabilities internally.
Repo map
| the spec — every decision, with rationale |
| meta-schema, Zod validator with teaching errors, path parser ( |
| renderer: rules engine, host bridge, single-file bundle ( |
| Cloudflare Worker: MCP handler, form Durable Objects ( |
| the installable plugin (connector + skill) |
| marketplace catalog pointing at |
Develop
pnpm install
pnpm -r test # 359 tests
pnpm --filter @mcpq/ui dev # offline host harness at localhost:5173/dev/
pnpm build # single-file renderer bundle, verified self-containedDeploying (packages/worker): pnpm exec wrangler deploy. Two secrets carry
the GitHub OAuth app's credentials —
wrangler secret put GITHUB_CLIENT_ID
wrangler secret put GITHUB_CLIENT_SECRET— from a GitHub OAuth app whose Authorization callback URL is
https://mcp-questionnaire.<account>.workers.dev/callback. Pipe secrets with
tr -d '\n': a trailing newline has bitten this deploy before. The allowlist
(GITHUB_ALLOWED_USERS) is a vars entry, not a secret, so it ships in
wrangler.jsonc and changes in a reviewable diff. .dev.vars holds obvious
fakes for local runs.
This server cannot be deployed
Maintenance
Related MCP Connectors
Your audit methodology inside Claude: findings, risks, controls and workpapers in your team format.
Turn PRDs and product ideas into structured specs so coding agents build your intent, not theirs.
Structured financial modeling for AI agents: build, version, audit models, export to Excel.
A structured financial modeling layer for AI agents. Build, version, and audit financial models without drift, then export to Excel, from Claude or any MCP client. Learn more: https://layerz.cc/for-agents
Related MCP Servers
- FlicenseNot gradedqualityCmaintenanceEnables AI agents to create visual, interactive interfaces with diagrams, structured input forms, and persistent state management instead of text-only responses.7-
- AlicenseBqualityCmaintenanceProvides AI systems with structured thinking frameworks and reasoning tools to maintain consistent problem-solving patterns across conversations. Enables multi-step reasoning, decision analysis, and systematic troubleshooting through invocable mental models.1516 npm19MIT
- AlicenseBqualityDmaintenanceProvides Claude with 44 tools for confidence gating, typed outputs, hallucination detection, and constraint enforcement during conversations.511MIT
- AlicenseNot gradedqualityDmaintenanceEnables AI agents to ask clarification questions and receive structured user input through a Human-in-the-Loop interface.1MIT