MCP SERVERHands a real form to a human: the model describes a quiz as JSON, the person fills it in at a local URL in their browser, and the answers come back as structured data.
LABSCO SUMMARY
Asking a human ten questions in a conversation gets you ten paragraphs to parse; here the same ten arrive as typed answers matched to their questions, including the ones left blank. Fetch the schema before building anything — the question types and their required fields are what the runner validates against, and the runner assigns the quiz id itself, so the definition should not carry one. Useful well beyond quizzes: structured interviews, intake forms, and any point where an agent needs a decision from a person in a form rather than in prose.
A way for a model to collect real input from a person. It serves the quiz as a web page on your machine, waits while the human answers, and returns each question paired with what was submitted.
Node.js 20 or newer, and a browser on the machine — the local server runs as npx -y @quiz-mcp/cli mcp and the person answering has to be at that machine to open the page. A hosted, OAuth-protected version runs on Cloudflare Workers if you would rather install nothing; it exposes the same tools plus per-user quiz management, and signs you in on the first call. On the hosted version, answers are stored in an eventually consistent store, so a read straight after a submission can come back empty — retry after a few seconds rather than treating it as lost.
One command — npx -y @quiz-mcp/cli mcp
USE CASES
Build the quiz to the schema, open it in their browser, and read back each question with its answer once they submit.
Describe the material and let the model generate questions across the supported types, then hand the URL to whoever is taking it.
Mix scales, long text and choice questions in one form, so responses come back comparable rather than as free prose.
Shut the runner down explicitly instead of leaving a local server listening after the work is finished.
Counted once with only startup setup. Other versions, modes or optional credentials may expose a different list.
get_answersReturns a compact pairing of each question with its submitted answer (or null if unanswered). Each question is projected to a minimal form: id, _kind, title, text, and the type-specific data needed to interpret the answer (options for choice types, pairs for match, items for sorting, min/max for scale, parts for fill_gaps, accept for upload). Use after the user has completed or partially completed a quiz started via start_quiz.
get_quiz_formatReturns the JSON Schema for a Quiz definition. MUST be called first before building a quiz via start_quiz — the schema describes all supported question types and their required fields. Note: the runner assigns the quiz id, so the definition itself does not include one.
start_quizRegisters a quiz in the runner and (optionally) opens it in the user's default browser. The runner assigns a fresh id; do not include one in the input. Returns the assigned quizId and the URL to open. The quiz must match the schema returned by get_quiz_format.
stop_runnerForcibly shuts down the HTTP runner-api server. Use this to explicitly end a quiz session before all quizzes are finished and read. No-op if the runner is already stopped.
BEFORE YOU INSTALL
4 tools when we connected. The hosted route, behind OAuth, adds get_quizzes and cleanup_quiz for six in total.
npx on PATH
Of the 7 documented setups, 6 run the server on your machine and 1 connects to a hosted endpoint.
CONTEXT COST
Rated by the size of the 4 tool definitions that ride along on every turn, not by how many there are
INSTALLATION
npx -y @quiz-mcp/cli mcp{
"mcpServers": {
"quiz-mcp": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"@quiz-mcp/cli",
"mcp"
]
}
}
}{
"mcpServers": {
"quiz-mcp-remote": {
"type": "http",
"url": "https://quiz-mcp-oauth.karerckor.workers.dev/mcp"
}
}
}{
"$schema": "https://opencode.ai/config.json",
"mcp": {
"quiz-mcp": {
"type": "local",
"command": [
"npx",
"-y",
"@quiz-mcp/cli",
"mcp"
],
"enabled": true
}
}
}{
"mcpServers": {
"quiz-mcp": {
"command": "npx",
"args": [
"-y",
"@quiz-mcp/cli",
"mcp"
]
}
}
}MCP servers
makloai
edgarrmondragon
TonisOrmisson
cyanheads