missed-deadline-mcp
A Korean-language loss-prevention MCP server that extracts deadlines from user-provided text, ranks them by risk, and generates actionable steps to avoid financial or opportunity loss.
README
missed-deadline-mcp (놓치면 손해 MCP)
A Korean-language, loss-prevention MCP server. Paste in the text you already have — a reservation notice, a coupon, a bill, an A/S warranty, or text copied from a screenshot/PDF — and it finds the deadlines that cost you money or opportunity if missed, ranks them by risk, and turns each into something you can actually do today.
It never crawls your accounts, never sends or cancels anything, and never invents a date it isn't sure about. Every extracted item comes back with a short source excerpt, a confidence score, and a "needs confirmation" flag when the text is ambiguous.

Quick start
Requires Bun.
bun install
bun run start
The server listens on http://localhost:3000 by default (override with PORT).
It exposes two endpoints:
POST /mcp— the MCP endpoint (streamable HTTP transport)GET /health— liveness check, returns{ "status": "ok", ... }
For hot-reload during development:
bun run dev
Tools
The server registers five MCP tools:
| Tool | What it does |
|---|---|
extract_loss_deadlines |
Extracts loss-bearing deadline items (refund, cancellation, payment, coupon, warranty, application…) from Korean text, a public URL, or fallback text copied from a screenshot/PDF. |
rank_loss_risk |
Sorts extracted items by deadline urgency, amount at stake, reversibility, and confidence. |
make_action_plan |
Drafts the first check / cancel / inquire / apply steps as safe guidance the user runs themselves. |
create_reminder_payload |
Produces a copy-pasteable reminder draft — no real calendar write. |
daily_loss_digest |
Bundles today's loss-risk items into a short Korean summary. |
Example: extract_loss_deadlines
Input shapes are validated with Zod. The most common call passes Korean text directly:
{
"sourceType": "text",
"sourceText": "주문 취소는 6월 25일까지 가능하며, 이후에는 환불 수수료 10%가 부과됩니다.",
"requestDate": "2026-06-19"
}
Other sourceType values: public_url (requires sourceUrl), and
screenshot / pdf (require fallbackText — the server processes copied text,
not raw binary files).
Configuration
| Env var | Default | Purpose |
|---|---|---|
PORT |
3000 |
HTTP port the server binds to. |
Copy .env.example to .env and adjust as needed (.env is gitignored).
The port can also be passed as --port <n> on the command line.
Development
bun run lint # biome check
bun run typecheck # tsc --noEmit
bun test # run the test suite
bun run build # bundle to dist/
Project structure
src/
index.ts # entry: starts the Bun HTTP server
server/ # Hono app, /health, port config
mcp/ # MCP server + tool schemas
tools/ # the five tool implementations
domain/ # contracts, Korean parsers, redaction, safe errors
ingest/ # public URL + text fallback ingestion, URL safety
storage/ # in-memory loss-deadline store
tests/ # unit + contract + endpoint tests (bun test)
docs/ # deployment, privacy, registration, smoke notes
deploy/ # Kakao Cloud manifest
Dockerfile # oven/bun runtime image
Deploy
A Dockerfile (based on oven/bun) builds a runtime image that serves the same
/mcp and /health endpoints. See docs/deployment.md for
the Kakao Cloud / PlayMCP registration path.
Privacy & scope
This is a preliminary hackathon MVP (PlayMCP). By design it:
- processes only user-supplied text, copied fallback text, or public URLs;
- does not crawl accounts, mailboxes, or SMS, and uses no private APIs;
- does not auto-pay, cancel, refund, apply, message, or write to a calendar;
- does not durably store raw input — masks phone/card/account-like strings in logs and errors.
It is not legal, medical, tax, or financial advice. See docs/privacy.md for details.
License
[License — not yet specified]
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。
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。
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。
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。
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。
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。
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。
mcp-server-qdrant
这个仓库展示了如何为向量搜索引擎 Qdrant 创建一个 MCP (Managed Control Plane) 服务器的示例。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。