DEV Community

Tawin Tangsukson
Tawin Tangsukson

Posted on

I built an MCP server that lets Claude read my study data — here's how

#education

I built an MCP server that lets Claude read my study data

tl;dr — Shiori is an open-source AI study app (React + Gemini). I added a Model Context Protocol server so you can ask Claude Code things like "what assignments are due this week?" and get real answers from your actual data.

The problem

I was using Claude Code to help me study, but it had no idea what I was actually studying. I'd have to paste my notes, deadlines, and grades into the chat every time.

What I wanted: Claude to already know my study context.

What Shiori does

Shiori is a full study companion: assignments, grades, GPA predictor, AI flashcard generation, spaced repetition, AI quiz generator, habit tracker, focus mode (Pomodoro), leaderboard, and a syllabus importer that extracts all assignments from a PDF/text dump.

It runs entirely in the browser — no server needed for the core features. Data lives in localStorage. You can try it at shiori-v1.vercel.app (click TRY DEMO, no account needed).

The MCP server

The interesting part: I added a /mcp server that exposes 6 tools:

get_study_summary     → overview of assignments, grades, streaks
get_assignments       → list with due dates, status, priority
get_grades            → GPA, course grades, trend
get_notes             → your markdown notes
add_assignment        → create assignment from Claude
get_flashcard_decks   → deck names + mastery %
Enter fullscreen mode Exit fullscreen mode

How it works

You export your data from Shiori (Settings → Export Data), which writes shiori-data.json. The MCP server reads that file and serves it to Claude.

// mcp/index.js (simplified)
server.tool('get_assignments', async () => {
  const data = JSON.parse(fs.readFileSync(DATA_PATH))
  const upcoming = data.assignments
    .filter(a => !a.completed)
    .sort((a, b) => a.dueDate - b.dueDate)
    .slice(0, 20)
  return { content: [{ type: 'text', text: JSON.stringify(upcoming, null, 2) }] }
})
Enter fullscreen mode Exit fullscreen mode

Claude Code setup

Add to .claude/mcp.json:

{
  "mcpServers": {
    "shiori": {
      "command": "node",
      "args": ["/path/to/Shiori-v1/mcp/index.js"],
      "env": { "SHIORI_DATA_PATH": "/path/to/shiori-data.json" }
    }
  }
}
Enter fullscreen mode Exit fullscreen mode

Then in Claude Code: what's due this week? → Claude calls get_assignments and gives you a real answer.

Claude Desktop setup

Same idea, add to claude_desktop_config.json:

{
  "mcpServers": {
    "shiori": {
      "command": "node",
      "args": ["/absolute/path/to/mcp/index.js"]
    }
  }
}
Enter fullscreen mode Exit fullscreen mode

Why this matters

The MCP pattern is powerful for personal tools. Your study data doesn't need to go through any API — it stays local. Claude reads the file directly. This is the "bring your own context" model taken seriously.

I plan to add more tools: create_study_plan, generate_quiz_from_notes, get_habit_streaks.

Try it

Stars appreciated — this is a solo project and GitHub discoverability matters for open source. PRs welcome — there are good-first issues open with exact file + line pointers.

Top comments (0)