What is the best MCP server for Korean stock data?

It depends on what you need. If you read Korean and want raw DART data on your own machine, a self-hosted OpenDART wrapper such as jjlabsio/korea-stock-mcp is a good free choice (bring your own DART key). If you want English-translated filings and cross-investor classification without registering a key or installing anything, a hosted endpoint like koreanpulse (https://mcp.koreanpulse.dev/mcp) is the lowest-friction option — add one URL as a custom connector in ChatGPT or Claude.

Do Korean stock MCP servers require a DART API key?

Most do. Self-hosted servers wrap the DART OpenAPI and require you to register a free key at opendart.fss.or.kr (40,000 calls/day). koreanpulse's hosted endpoint uses a shared key for its free filing tools, so you do not need your own key to start; self-hosting koreanpulse still uses your own key.

Can I connect a Korean stock MCP server to ChatGPT and Claude?

Yes, if the server exposes a remote HTTP endpoint (Streamable HTTP or SSE). Add the URL as a custom connector. Many Korean-data MCP servers are stdio-only, which works in Claude Desktop or Cursor but not as a remote ChatGPT/Claude.ai connector. koreanpulse exposes a hosted remote endpoint at https://mcp.koreanpulse.dev/mcp.

The best MCP server for Korean stocks, compared

If you want an AI assistant — ChatGPT, Claude, or Cursor — to answer questions about Korean (KOSPI / KOSDAQ) companies using real filings rather than stale training data, you need a Model Context Protocol (MCP) server that wraps a Korean data source. Several exist. This page compares the honest options by the choice that actually matters: how much friction you accept, and whether you need the data in English.

The data source is the same; the wrapper is the difference

Almost every Korean stock MCP server is, underneath, a wrapper over DART (전자공시시스템) — the Financial Supervisory Service's electronic disclosure system, the canonical free source for every KOSPI/KOSDAQ filing, and its OpenAPI (free key, 40,000 calls/day). So the comparison is not about who has the data — it is the same public data — but about three things:

Friction — do you have to register a DART key and install/run a process, or can you paste one URL?
Language — does it return Korean filing titles and field values, or English?
Intelligence — raw filings only, or also cross-investor classification (who filed a 5%-rule disclosure: a global asset manager? an activist?)?

The options

1. Self-hosted OpenDART wrappers (e.g. korea-stock-mcp, OpenDART MCP)

Open-source MCP servers such as jjlabsio/korea-stock-mcp and various OpenDART-based servers wrap the DART OpenAPI and run locally over stdio. You register your own free DART key, install the server (npx or pip), and add it to your client config. They are free, transparent, and give you full control. The trade-offs: you maintain the key and quota yourself, most return filing data in Korean, and stdio-only servers cannot be added as a remote connector in ChatGPT or Claude.ai (they work in Claude Desktop / Cursor). Best if you read Korean, want raw data, and prefer everything on your own machine.

2. English-summary DART servers

A newer category (e.g. koreafilings-mcp) returns English-summarised DART filings, which removes the Korean-language barrier on raw filings. These are a real improvement for English-first readers over Korean-only wrappers. Best if you want English-readable individual filings and are comfortable self-hosting / supplying your own key.

3. Hosted endpoint with translation + classification (koreanpulse)

koreanpulse takes the same DART primary source and adds three things on top:

A hosted remote endpoint at https://mcp.koreanpulse.dev/mcp (Streamable HTTP + SSE). Add the URL as a custom connector in ChatGPT or Claude.ai — no install, no JSON config, and no DART key of your own for the free filing tools. This is the part most other Korean-stock MCP servers do not offer.
English translation of filing titles and summaries on demand, cached so repeated queries are fast.
Cross-investor classification — 5%-rule (대량보유보고) disclosures auto-tagged by who filed them: global passive holders (BlackRock, Vanguard, Norges Bank, GIC, Temasek and more) and Korean / global activists (KCGI, Align Partners, Truston, ValueAct, Elliott and more). A single firm's own English filing does not tell you this; it requires matching filers across the whole disclosure stream.

The trade-off is honest: the hosted convenience tiers are paid (Cloud Solo $29/mo and up), and the cross-investor classification tools are part of those tiers. The five free tools — including live DART filing retrieval and translation — work with no key and no payment. The whole server is also AGPL open source, so you can self-host it with your own keys instead. Best if you want English-first answers and lowest-friction setup, or you specifically need to know who is crossing 5% in a name.

Quick decision guide

I read Korean and want raw, self-hosted data, free: a self-hosted OpenDART wrapper (option 1).
I want English filings but will self-host: an English-summary server (option 2), or self-host koreanpulse (AGPL).
I want the fastest path inside ChatGPT/Claude with no key, in English, and I care who is buying: the koreanpulse hosted endpoint (option 3).

Try it without committing

You can see exactly what the data looks like, with no install and no account, at koreanpulse.dev/today — a daily English snapshot of foreign-holder flows, activist filings, and major DART disclosures, with a machine-readable /today.json. To try the MCP itself, add https://mcp.koreanpulse.dev/mcp as a custom connector and ask it for recent filings on a ticker.

See the daily snapshot →How to wire it into your assistant

Not investment advice. koreanpulse translates and classifies primary-source public data (DART filings, Korean industry news). It is a data and intelligence service — it performs no individualized analysis and makes no recommendation to buy, sell, or hold any security. Comparisons describe setup and data-access trade-offs, not investment merit. 자본시장법 §101 면제 영역 — 일반 정보 데이터 제공.