🧠 Simple MCP Server with Node.js

This is a simple prototype of a Model Context Protocol (MCP) server built with Node.js and Express.
It interacts with OpenAI's API to simulate a context-aware assistant—for example, one that helps write stories.

---

🚀 What You Can Do

- Structure your context with system, user, steps, and resources
- Call OpenAI API to generate story plots using the context
- Add characters dynamically via API
- See how changing context affects LLM responses

---

📦 Requirements

- Node.js v18+
- npm
- OpenAI API key (Get one here)

---

🛠 Setup

1. Clone this repository

git clone https://github.com/takehisa10098/plot-mcp-server.git
cd plot-mcp-server

2. Install dependencies

npm install

3. Create .env file

OPENAI_API_KEY=sk-xxxxxxxxxxxxxxxxxxxxxxxx

4. Create context.json file

{
  "system": "あなたは小説執筆のサポートAIです。",
  "user": {
    "goal": "和風ファンタジーの短編小説を書きたい",
    "constraints": ["文字数は3000字以内", "テーマは『喪失と再生』"]
  },
  "steps": [
    { "name": "キャラクター作成", "status": "done" },
    { "name": "プロット生成", "status": "in_progress" }
  ],
  "resources": {
    "characters": [
      { "name": "アカネ", "role": "主人公", "trait": "寡黙で芯が強い" }
    ],
    "world": "戦国時代風の異世界。陰陽術が存在する。"
  }
}

5. Create server.js

const express = require('express');
const fs = require('fs');
const dotenv = require('dotenv');
const { getPlotSuggestion } = require('./openai');
dotenv.config();

const app = express();

app.use(express.json());
const PORT = 3000;

const CONTEXT_PATH = './context.json';
const readContext = () => JSON.parse(fs.readFileSync(CONTEXT_PATH, 'utf-8'));
app.get('/context', (req, res) => {

  const context = readContext();

  res.json(context);

});
app.post('/step/plot', async (req, res) => {

  const context = readContext();

  const plot = await getPlotSuggestion(context);

  res.json({ plot });

});
app.post('/update/characters', (req, res) => {

  const context = readContext();

  const newCharacter = req.body;
if (!newCharacter.name || !newCharacter.role || !newCharacter.trait) {

    return res.status(400).json({ error: 'name, role, and trait are required' });

  }
context.resources.characters.push(newCharacter);

  fs.writeFileSync(CONTEXT_PATH, JSON.stringify(context, null, 2), 'utf-8');

  res.json({ message: 'Character added', character: newCharacter });

});
app.listen(PORT, () => {

  console.log(MCP server running at http://localhost:${PORT});

});

6. Create openai.js

const { OpenAI } = require('openai');
require('dotenv').config();
const openai = new OpenAI({

  apiKey: process.env.OPENAI_API_KEY,

});
async function getPlotSuggestion(context) {

  const { system, user, resources } = context;
const messages = [

    { role: 'system', content: system },

    {

      role: 'user',

      content: 

目標: ${user.goal}

制約: ${user.constraints.join(', ')}

キャラクター: ${JSON.stringify(resources.characters)}

世界観: ${resources.world}
これを踏まえて、小説のプロット案を出してください。

      

    }

  ];
const res = await openai.chat.completions.create({

    model: 'gpt-4-turbo',

    messages,

  });
return res.choices[0].message.content;

}
module.exports = { getPlotSuggestion };

---

🧪 How to Use

Start the server

node server.js

Get current context

curl http://localhost:3000/context

Generate a plot

curl -X POST http://localhost:3000/step/plot

Add a new character

curl -X POST http://localhost:3000/update/characters \
  -H "Content-Type: application/json" \
  -d '{
    "name": "ユキ",
    "role": "謎の旅人",
    "trait": "静かで何かを知っているような雰囲気"
  }'

---

💡 What is MCP?

Model Context Protocol (MCP) is a proposed format for structuring contextual information for LLMs.
It allows applications to define system, user, steps, and resources that represent the ongoing state of a conversation or project.

This project shows how you can design an LLM interaction by modifying structured context, instead of rewriting your prompt each time.

---

🔗 Related

- Anthropic MCP Docs
- OpenAI Function Calling

---