Richgo Finance MCP
Enables searching and analyzing Korean stocks using Richgo Finance data, including stock details, rankings, undervaluation, AI rankings, and curation screens.
README
Richgo Finance MCP
Model Context Protocol (MCP) server for public data exposed by
finance.richgo.ai.
It lets MCP clients search Korean stocks, fetch Richgo stock detail data, and use Richgo's ranking, undervaluation, AI ranking, and curation screens from an assistant.
This project only calls public Richgo Finance endpoints by default. It does not bypass authentication or scrape private account data.
Tools
richgo_search_stock: search by Korean stock name or ticker.richgo_get_stock_detail: fetch public detail for a ticker.richgo_analyze_stock: search/alias resolve, then fetch compact analysis.richgo_get_scores: score ranking table.richgo_get_undervalued: undervalued stock tabs.richgo_get_ai_rankings: AI ranking tab.richgo_get_curation: curated screener buckets.richgo_get_breakout: breakout/special signal endpoint.richgo_get_market_api: generic/api/market/...wrapper.richgo_get_public_api: constrained generic/api/...wrapper.richgo_get_consensus_picks: merge Richgo scores, undervaluation, gap, and AI rankings into one ranked candidate list.richgo_build_portfolio: build an integer-share portfolio from Richgo consensus signals for a budget and risk profile.richgo_save_obsidian_report: save a generated portfolio report to a user-provided Obsidian vault path.richgo_guided_portfolio: beginner-friendly portfolio wizard that uses MCP form elicitation when the client supports it.richgo_service_catalog: list exposed services.
Install
git clone https://github.com/reallygood83/rfmcp.git
cd rfmcp
npm install
npm run build
Run Locally
npm start
On the original author's Mac, Homebrew Node was broken, so this bundled Node path was used:
export PATH=/Users/moon/.cache/codex-runtimes/codex-primary-runtime/dependencies/node/bin:$PATH
npm run build
npm start
Codex or Claude MCP config
{
"mcpServers": {
"richgo-finance": {
"command": "node",
"args": ["/absolute/path/to/rfmcp/dist/index.js"]
}
}
}
Example Prompts
You do not need to know the MCP tool names. Ask naturally:
아이티센글로벌 리치고 데이터로 분석해줘/rf 셀트리온 분석해줘삼성전자와 SK하이닉스 리치고 상세 지표 비교해줘리치고 저평가 total 상위 10개 보여줘AI 랭킹 상위 종목을 요약해줘리치고 컨센서스 기준으로 오늘 투자 매력 높은 종목 10개 뽑아줘500만원 중립형 포트폴리오 만들어줘초보자용 리치고 포트폴리오 마법사 시작해줘나는 잘 모르니까 질문하면서 300만원 투자 전략 만들어줘오늘 리치고 기준으로 뭐가 제일 좋아?1000만원이면 몇 종목을 몇 주씩 사면 좋을까?이 결과를 내 Obsidian 볼트에 보고서로 저장해줘
Natural Language Routing
Most MCP clients choose tools from the descriptions below. These are the intended natural-language routes:
| User says | Preferred tool | What happens |
|---|---|---|
삼성전자 분석해줘, 005930 리치고로 봐줘 |
richgo_analyze_stock |
Resolves the stock and returns compact Richgo detail. |
저평가 상위 10개, AI 랭킹 보여줘 |
richgo_get_undervalued / richgo_get_ai_rankings |
Returns the requested screen. |
오늘 투자 매력 높은 종목, 뭐가 좋아?, 컨센서스 픽 |
richgo_get_consensus_picks |
Combines scores, undervaluation, gap, and AI ranking into one candidate list. |
500만원 포트폴리오, 몇 주씩 살까?, 예산으로 전략 짜줘 |
richgo_build_portfolio |
Builds an integer-share portfolio with cash balance and risks. |
초보자용, 질문하면서 도와줘, 마법사 시작 |
richgo_guided_portfolio |
Uses form elicitation when supported, then builds the portfolio. |
Obsidian에 저장, 볼트에 보고서로 저장 |
richgo_save_obsidian_report |
Saves only when a vault path is provided or configured privately. |
Optional Slash Command
This repo includes a Claude-style slash command template at commands/rf.md.
Install it as /rf with:
mkdir -p ~/.claude/commands
cp commands/rf.md ~/.claude/commands/rf.md
After that, use prompts like:
/rf 셀트리온 분석해줘
The /rf command is tuned to produce data-backed Korean reports, not just a
metric dump. It should connect revenue, profitability, cash flow, financial
safety, valuation, and Richgo percentile signals into a concrete story that is
easy to understand.
Portfolio and Obsidian reports
The portfolio tools intentionally do not hardcode a personal vault path. This keeps the published MCP reusable for other users.
Recommended natural-language flow:
초보자용 리치고 포트폴리오 마법사 시작해줘
If your MCP client supports form elicitation, the server asks for budget, risk profile, number of stocks, and whether to save to Obsidian. If the client does not support elicitation, call:
500만원 중립형 포트폴리오 만들어줘
To save Obsidian reports, either pass vaultPath in the tool call:
{
"budget": 5000000,
"riskProfile": "balanced",
"vaultPath": "/absolute/path/to/your/vault",
"noteDir": "Richgo"
}
Or keep the path in your private shell/MCP environment:
export RICHGO_OBSIDIAN_VAULT_PATH='/absolute/path/to/your/vault'
Do not commit personal vault paths to this repository.
Optional auth
Public endpoints work without auth. If you legitimately need login-only Richgo features, set your own cookie:
export RICHGO_COOKIE='...'
This server does not bypass authentication.
Disclaimer
Richgo Finance data is informational only and is not investment advice. Final investment decisions and responsibility belong to the user.
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。