` tags, `{\ wikitable` |
| 23 | `convert_to_bbcode` | BBCode | `[b]`, `[i]`, `[code]`, `[url]`, `[img]` tags for phpBB/vBulletin forums |
| 24 | `convert_to_textile` | Textile | Markup for Redmine, Basecamp, and CMS platforms |
| 25 | `convert_to_orgmode` | Emacs Org Mode | `*` headers, `#+BEGIN_SRC`/`#+END_SRC` code, `[[url][text]]` links |
### Import Tools (1)
| # | Tool | Description |
|---|---|---|
| 26 | `html_to_markdown` | Convert HTML content (full document or fragment) back to Markdown |
### Repair & Lint Tools (2)
| # | Tool | Description |
|---|---|---|
| 27 | `repair_markdown` | Fix broken Markdown from LLM output or copy-paste — unclosed code fences, broken tables, stray emphasis markers, missing heading spaces, broken links |
| 28 | `lint_markdown` | Lint and report issues as JSON array with line number, severity, rule name, message, and fixable flag |
### Document Analysis Tools (5)
| # | Tool | Description |
|---|---|---|
| 29 | `extract_code_blocks` | Extract all code blocks with language, content, and start/end line numbers |
| 30 | `extract_links` | Extract all links and images with text, URL, line number, and type (inline, reference, image, autolink) |
| 31 | `generate_toc` | Generate Markdown Table of Contents from headings with configurable `max_depth` (1–6) |
| 32 | `analyze_document` | Comprehensive statistics — line/word/character/paragraph/sentence counts, heading/code/table/link/image counts, estimated reading time |
| 33 | `extract_structure` | Bird's-eye document architecture — statistics, heading outline, code block summary, link summary |
### Batch Conversion (1)
| # | Tool | Input | Description |
|---|---|---|---|
| 34 | `batch_convert` | `items[]` × `formats[]` | Batch-convert multiple Markdown documents to 22 formats in one call. Error-isolated per item+format. Optional `output_dir` for file output. Formats: txt, csv, json, xml, xlsx, latex, rtf, docx, pdf, html, md, email, slack, discord, jira, confluence, asciidoc, rst, mediawiki, bbcode, textile, orgmode |
---
## 6 Built-in Prompts
Pre-configured prompt templates that orchestrate multiple tools:
| Prompt | Description |
|---|---|
| `convert-document` | Convert Markdown to any supported format (PDF, DOCX, HTML, LaTeX, CSV, JSON, XML, XLSX, RTF, PNG, TXT, MD) |
| `extract-tables` | Extract all tables from Markdown and export as CSV or XLSX |
| `format-for-sharing` | Harmonize formatting, then convert to both PDF and HTML for distribution |
| `analyze-and-repair` | Lint → Repair → Re-lint to confirm all issues resolved |
| `convert-for-platform` | Convert Markdown to any platform format (Slack, Discord, JIRA, Confluence, AsciiDoc, RST, MediaWiki, BBCode, Textile, Org Mode) |
| `document-overview` | Full overview — statistics, TOC, code blocks, and links in one report |
---
## 2 Resources
| Resource URI | Description |
|---|---|
| `markdown-formatter://supported-formats` | Complete JSON list of all 23+ output formats with tool names and types |
| `markdown-formatter://conversion-guide` | Guide for choosing the right output format based on your use case |
---
## Quick Start
### Install via npx (Recommended)
No installation needed — just configure your MCP client:
```json
{
"mcpServers": {
"ai-answer-copier": {
"command": "npx",
"args": ["-y", "@xjtlumedia/markdown-mcp-server"]
}
}
}
```
### Install Globally
```bash
npm install -g @xjtlumedia/markdown-mcp-server
```
Then configure:
```json
{
"mcpServers": {
"ai-answer-copier": {
"command": "markdown-mcp-server"
}
}
}
```
---
## Configuration by AI Client
### 🔌 Claude Desktop
Edit `%APPDATA%\Claude\claude_desktop_config.json` (Windows) or `~/Library/Application Support/Claude/claude_desktop_config.json` (macOS):
```json
{
"mcpServers": {
"ai-answer-copier": {
"command": "npx",
"args": ["-y", "@xjtlumedia/markdown-mcp-server"]
}
}
}
```
Restart Claude Desktop. You'll see a 🔌 icon — all 34 tools are now available.
### VS Code (GitHub Copilot)
Add to `.vscode/mcp.json` or VS Code settings:
```json
{
"servers": {
"ai-answer-copier": {
"command": "npx",
"args": ["-y", "@xjtlumedia/markdown-mcp-server"]
}
}
}
```
### Cursor / Windsurf / Any MCP Client
Use the same `npx` command pattern above in your client's MCP configuration.
### DIY version (local build)
```json
{
"mcpServers": {
"markdown-formatter": {
"command": "node",
"args": [
"YOUR_PATH_TO_PROJECT/mcp-server/dist/mcp-server/src/index.js"
]
}
}
}
```
*Replace `YOUR_PATH_TO_PROJECT` with the absolute path to this folder.*
### HTTP Mode (Vercel)
A hosted HTTP endpoint is available for browser-based and remote integrations:
```
https://ai-answer-copier.vercel.app/api/mcp
```
Test with MCP Inspector:
```bash
npx @modelcontextprotocol/inspector https://ai-answer-copier.vercel.app/api/mcp
```
---
## Usage Examples
Once connected, just talk to your AI naturally:
> **"Batch-convert 5 markdown documents into DOCX, PDF, and HTML simultaneously and save them to my Downloads folder."**
> → AI uses `batch_convert` with items[] and formats[]
> **"Generate 10 physics questions about Newton's Laws and export them as a Kahoot CSV."**
> → AI uses `convert_to_csv`
> **"Convert this markdown into a Word document and save it to my Desktop."**
> → AI calls `convert_to_docx` with `output_path`
> **"Take these lecture notes and produce a print-ready PDF."**
> → AI calls `convert_to_pdf` with Puppeteer rendering, KaTeX math, and syntax highlighting
> **"Turn this into a Moodle-compatible XML quiz bank."**
> → AI calls `convert_to_xml`
> **"Format this for Slack and send it to the team channel."**
> → AI calls `convert_to_slack`
> **"Check this Markdown for issues and fix them."**
> → AI calls `lint_markdown` then `repair_markdown`
> **"Give me an overview of this document — word count, structure, all links."**
> → AI chains `analyze_document` + `generate_toc` + `extract_links`
---
## Features
### 22 Output Formats
Export to PDF, DOCX, HTML, LaTeX, CSV, JSON, XML, XLSX, RTF, PNG, TXT, Markdown — plus Slack, Discord, JIRA, Confluence, AsciiDoc, RST, MediaWiki, BBCode, Textile, Org Mode, and Email HTML.
### Batch Conversion
Convert multiple Markdown documents to multiple formats in one `batch_convert` call. Errors are isolated per item+format so one failure never blocks the rest.
### Math & Code as First-Class Citizens
Full support for KaTeX math expressions and syntax-highlighted code blocks. Your `$\sqrt{x^2 + y^2}$` and Python snippets survive every conversion perfectly.
### Smart Binary Handling
Binary formats (PDF, DOCX, XLSX, PNG) don't dump raw base64 into chat. The server returns actionable guidance so the AI knows to save to a file path.
### Document Analysis & Repair
Lint Markdown, auto-repair broken formatting, extract code blocks and links, generate Table of Contents, and get comprehensive document statistics — all without leaving the chat.
### GFM (GitHub Flavored Markdown)
Tables, task lists, strikethrough, autolinks — all parsed correctly via `remark-gfm`.
### Puppeteer-Powered PDF & PNG
PDF and image exports use headless Chromium for pixel-perfect rendering with full CSS styling, KaTeX math, syntax highlighting, and print-optimized layouts.
### Self-Contained
Zero runtime dependencies on external APIs. Everything runs locally. Your data never leaves your computer.
---
## Architecture
```
┌─────────────────────────┐
│ AI Client │
│ (Claude, Copilot, │
│ Cursor, Windsurf) │
└──────────┬──────────────┘
│ MCP Protocol (stdio / HTTP)
▼
┌─────────────────────────┐
│ AI Answer Copier │
│ MCP Server (34 tools) │
│ │
│ ┌── Conversion ────────┐│
│ │ remark/rehype, docx, ││
│ │ puppeteer, xlsx, ││
│ │ CSV/JSON/XML/RTF/ ││
│ │ LaTeX/TXT parsers ││
│ └──────────────────────┘│
│ ┌── Platform ──────────┐│
│ │ Slack, Discord, JIRA,││
│ │ Confluence, AsciiDoc,││
│ │ RST, MediaWiki, ││
│ │ BBCode, Textile, Org ││
│ └──────────────────────┘│
│ ┌── Batch ─────────────┐│
│ │ items[] × formats[] ││
│ │ 22 formats, isolated ││
│ └──────────────────────┘│
│ ┌── Analysis ──────────┐│
│ │ Code blocks, links, ││
│ │ TOC, stats, structure││
│ └──────────────────────┘│
│ ┌── Repair ────────────┐│
│ │ Lint, auto-repair ││
│ └──────────────────────┘│
│ ┌── Import ────────────┐│
│ │ HTML → Markdown ││
│ └──────────────────────┘│
└─────────────────────────┘
```
---
## What's New in v2.1.0
This release hardens both the local (`npx`) and remote Vercel MCP transports with 8 logic fixes:
| # | Fix | Area |
|---|---|---|
| 1 | **Browser not closed on error** — `try/finally` around `browser.close()` so Chromium never leaks on PDF/PNG failures | Both transports |
| 2 | **`generate_html` missing KaTeX CSS** — restored viewport meta, KaTeX stylesheet (SRI hash), and full inline styles so math renders correctly | Vercel `/api/mcp` |
| 3 | **`convert_to_html` incomplete styles** — added SRI integrity hash, table borders, and blockquote styling to match local output | Vercel `/api/mcp` |
| 4 | **Session map memory leak** — `instances` Map now evicts stale sessions after 30 minutes (`SESSION_TTL_MS`); `McpInstance` tracks `lastUsed` | Vercel `/api/mcp` |
| 5 | **`keepAlive` interval not cleared** on normal SSE stream end — added inner `try/finally` that always calls `clearInterval(keepAlive)` | Vercel `/api/mcp` |
| 6 | **`_initialized` SDK-internal access** — replaced bare `@ts-ignore` with a `typeof` guard so the server fails gracefully after SDK upgrades | Vercel `/api/mcp` |
| 7 | **No input size validation** — added 1 GB hard limit (`MAX_INPUT_BYTES`) to all tool calls to prevent runaway CPU/memory on oversized payloads | Both transports |
| 8 | **DELETE method creates new session** — handler now returns `200` immediately on DELETE, preventing phantom session creation | Vercel `/api/mcp` |
---
## Known Limitations
- **PDF / PNG require local Chromium** — `convert_to_pdf` and `convert_to_image` launch a headless browser. This works fine with `npx` but is unavailable on Vercel's free tier (no bundled Chromium). Set the `PUPPETEER_EXECUTABLE_PATH` environment variable to point to your local Chrome if auto-detection fails.
- **Chromium version on Vercel** — the remote endpoint uses `@sparticuz/chromium-min` with a pinned tar at v131.0.1. If Vercel bumps the Lambda runtime you may need to update the `CHROMIUM_TAR_URL` env var.
- **1 MB input limit** — all tool calls reject inputs larger than 1 MB. Split large documents into sections before converting.
- **Session TTL (remote only)** — sessions on the Vercel endpoint expire after 30 minutes of inactivity. Long-running stateful workflows should issue periodic keep-alive calls.
---
## Development
```bash
# Clone the repository
git clone https://github.com/XJTLUmedia/AI_answer_copier.git
# Web app
npm install
npm run dev
# MCP server
cd mcp-server
npm install
npm run dev # hot reload
npm run build # production build
npm run inspector # test with MCP Inspector
```
---
## Tech Stack
| Component | Technology |
|---|---|
| MCP SDK | `@modelcontextprotocol/sdk` |
| Markdown Parser | `unified` + `remark` + `rehype` |
| Math Rendering | `remark-math` + `rehype-katex` |
| Word Export | `docx` |
| Excel Export | `xlsx` |
| PDF/PNG Export | `puppeteer` |
| Schema Validation | `zod` |
| Runtime | Node.js (ESM) |
---
## Contributing
We welcome contributions! Whether it's a new export format, a bug fix, or documentation improvement — please consider making a pull request.
1. Fork the [repository](https://github.com/XJTLUmedia/AI_answer_copier)
2. Create a feature branch
3. Submit a pull request
---
## License
[MIT](https://opensource.org/licenses/MIT) © [XJTLUmedia](https://github.com/XJTLUmedia)
---
Built by educators, for educators.
Reclaim 5 hours of your week. Let the AI handle the formatting.