One Answer

Many answers in. One answer out.

One Answer is a tiny MCP/API primitive for agents that ask multiple models, prompts, or runtimes and need one decision-ready result.

npx --yes --package @hyena0x/one-answer one-answer-mcp

The Problem

Multi-model products are easy to demo and hard to trust.

Users do not want five model transcripts. They want to know:

• what should I do?
• what do the answers agree on?
• where could this still be wrong?

One Answer packages that into one narrow tool.

The Primitive

one_answer(question, preset) -> answer + caveats + confidence

Use it when a product, agent, or workflow needs a final answer instead of a comparison table.

What It Returns

• final_answer: the answer you can show or act on
• consensus_points: what the answer paths agree on
• divergence_points: meaningful disagreement
• uncertainties: what is still unresolved
• confidence: level plus reason

Quick Start

npm install @hyena0x/one-answer

Add One Answer to any MCP host that supports stdio servers:

{
  "mcpServers": {
    "one-answer": {
      "command": "npx",
      "args": ["--yes", "--package", "@hyena0x/one-answer", "one-answer-mcp"]
    }
  }
}

Then call:

one_answer

Library Usage

import { runOneAnswer } from "@hyena0x/one-answer";

const result = await runOneAnswer({
  question: "Should this product stay MCP-first?",
  preset: "deep-reasoning",
  goal: "decision",
  audience: "developer",
});

console.log(result.final_answer);

Output Example

{
  "final_answer": "Use the narrow MCP-first path first.",
  "consensus_points": ["The smaller surface is easier to adopt."],
  "divergence_points": ["A thin UI may still help demos later."],
  "uncertainties": ["Real adoption still needs validation."],
  "confidence": {
    "level": "medium",
    "reason": "The answer paths agree on sequence, but usage data is still missing."
  }
}

Local Smoke

Default local runs use a built-in stub, so you can verify the package without calling a model provider.

npm run synthesize -- ./examples/request.decision.json

Where It Fits

• coding agents deciding between implementation paths
• research assistants reconciling model opinions
• product workflows that need a single recommendation
• internal tools that need confidence without a compare UI

V1 Promise

One public tool:

one_answer

No chat UI. No dashboard. No storage. No product ceremony.

Use a Real Provider

Real OpenAI-compatible runtime setup, MCP host env config, and local demos live in the usage guide.

• Usage guide

Docs

• Input schema
• Output schema
• Error schema
• Preset semantics
• Prompt contract
• Golden eval
• Changelog