SpecDown MCP Server
Gives AI assistants direct read/write access to SpecDown Markdown spec documents, enabling spec-driven development without copy-pasting.
README
SpecDown MCP Server — AI-Native Markdown MCP for AI Assistants
<p align="center"> <img src="https://img.shields.io/npm/v/specdown-mcp?color=blue" alt="npm version" /> <img src="https://img.shields.io/node/v/specdown-mcp" alt="node" /> <img src="https://img.shields.io/npm/dm/specdown-mcp" alt="downloads" /> <img src="https://img.shields.io/badge/MCP-compatible-green" alt="MCP compatible" /> <img src="https://img.shields.io/badge/spec--as--code-MCP-blueviolet" alt="spec as code" /> </p>
AI-native Markdown MCP server for SpecDown — give Claude, Cursor, Copilot, and other AI coding assistants direct read/write access to your Markdown spec documents.
Stop copy-pasting specs into AI chat. Connect once, AI reads your Markdown docs forever — AI-native spec-driven development.
Jump to your IDE: Cursor • Claude Desktop • Claude Code • Windsurf • OpenCode • VS Code • Codex CLI
What is AI-Native Markdown MCP?
Model Context Protocol (MCP) is an open standard that lets AI assistants connect to external tools and data sources. specdown-mcp is an AI-native Markdown MCP server that exposes your SpecDown spec documents as MCP resources — so Claude, Cursor, and other AI tools can read, search, and edit your specs directly.
AI-native Spec as Code + AI = spec-driven development:
Engineer writes Markdown spec → AI reads spec via MCP → AI implements feature from spec
Features
- Read & search Markdown docs — list projects, browse document tree, full-text search
- Edit docs — update Markdown content, create new documents (auto-versioned)
- Upload images — attach image assets and get markdown links back for documents
- Inline comments — add and list comments anchored to specific text
- Sync planning for agents — inspect remote snapshots, plan local-vs-remote sync, and apply remote sync operations
- AI-native spec-driven development — give AI full Markdown context with zero copy-paste
- 13 MCP tools — complete read/write and sync-planning API for your spec documents
- MCP resources —
specdown://projects,specdown://project/{id}
Prerequisites
- SpecDown account — free plan available
- API key — generate at Settings → API Keys
Quick Start
No install needed — npx runs it on demand:
# Verify it works
SPECDOWN_API_KEY=your_key npx specdown-mcp
IDE Configuration
Replace YOUR_API_KEY with your key from SpecDown Settings.
Cursor
Config path: .cursor/mcp.json (project) or ~/.cursor/mcp.json (global)
{
"mcpServers": {
"specdown": {
"command": "npx",
"args": ["-y", "specdown-mcp"],
"env": {
"SPECDOWN_API_KEY": "YOUR_API_KEY"
}
}
}
}
Claude Desktop
macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
Windows: %APPDATA%\Claude\claude_desktop_config.json
{
"mcpServers": {
"specdown": {
"command": "npx",
"args": ["-y", "specdown-mcp"],
"env": {
"SPECDOWN_API_KEY": "YOUR_API_KEY"
}
}
}
}
Claude Code
{
"mcpServers": {
"specdown": {
"command": "npx",
"args": ["-y", "specdown-mcp"],
"env": {
"SPECDOWN_API_KEY": "YOUR_API_KEY"
}
}
}
}
Windsurf
Config path: ~/.windsurf/mcp.json or .windsurf/mcp.json (project)
{
"mcpServers": {
"specdown": {
"command": "npx",
"args": ["-y", "specdown-mcp"],
"env": {
"SPECDOWN_API_KEY": "YOUR_API_KEY"
}
}
}
}
OpenCode
Config path: opencode.json (project root) or ~/.config/opencode/opencode.json
{
"mcpServers": {
"specdown": {
"command": "npx",
"args": ["-y", "specdown-mcp"],
"env": {
"SPECDOWN_API_KEY": "YOUR_API_KEY"
}
}
}
}
VS Code
Config path: .vscode/mcp.json
{
"servers": {
"specdown": {
"type": "stdio",
"command": "npx",
"args": ["-y", "specdown-mcp"],
"env": {
"SPECDOWN_API_KEY": "YOUR_API_KEY"
}
}
}
}
Codex CLI
Config path: ~/.codex/config.toml
[mcp_servers.specdown]
command = "npx"
args = ["-y", "specdown-mcp"]
[mcp_servers.specdown.env]
SPECDOWN_API_KEY = "YOUR_API_KEY"
Available MCP Tools
| Tool | Description |
|---|---|
list_projects |
List all projects you have access to |
list_documents |
List Markdown documents in a project |
list_project_files |
List preview-only project attachments |
read_document |
Read full Markdown content by ID or project+path |
read_project_file |
Read attachment metadata, text preview, embed ref, and download URL |
search_documents |
Full-text search across all Markdown spec docs |
read_project_context |
Get project overview: tree, README, description |
list_comments |
List inline comments on a document |
add_comment |
Add a comment (anchored to text or threaded reply) |
create_document |
Create a new Markdown document or folder |
update_document |
Replace Markdown content (auto-versioned on change) |
upload_image |
Upload an image asset and return a markdown link |
upload_project_file |
Upload HTML/HTM as editable documents, or other attachments with a [@/path] embed reference |
get_sync_status |
Return remote sync snapshots and summary for a project subtree |
plan_sync |
Compare local snapshots with remote docs and produce a sync plan |
apply_sync_plan |
Apply remote upsert/delete operations produced from a sync plan |
Environment Variables
| Variable | Required | Description |
|---|---|---|
SPECDOWN_API_KEY |
Yes | Your API key from SpecDown Settings |
SPECDOWN_API_URL |
No | Override API base URL (for self-hosted instances) |
Troubleshooting
Error: SPECDOWN_API_KEY must be set
→ Add SPECDOWN_API_KEY to the env block in your IDE's MCP config.
Unauthorized or 401
→ Key may be invalid or expired. Regenerate at Settings → API Keys.
Server not appearing in IDE
→ Restart the IDE after editing MCP config. Cursor: Cmd+Shift+P → "MCP: Reload Servers".
npx: command not found
→ Install Node.js ≥ 18 from nodejs.org.
Self-hosted SpecDown
Point the MCP server at your own instance:
"env": {
"SPECDOWN_API_KEY": "YOUR_API_KEY",
"SPECDOWN_API_URL": "https://your-specdown.example.com"
}
Related
- SpecDown — Markdown editor online, Spec as Code platform
- specdown-cli — Markdown CLI for terminal, CI/CD, and automation
- Get API Key
- Report issue
License
MIT
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。