korea-scholarship-mcp

A FastMCP stdio server exposing two Korean bibliographic services — the Korea Citation Index (KCI, 한국학술지인용색인, National Research Foundation of Korea) and Open Access Korea (OAK, 오픈액세스코리아, National Library of Korea) — as eight tools for Claude Desktop and other MCP clients.

It is the Korean counterpart to cinii-mcp and jstage-mcp and returns the same response envelope, so the three can be read side by side in trilateral work.

What this is for

Korean-language scholarship, through the Korea Citation Index and Open Access Korea.

Search KCI for articles in Korean-registered journals; pull a full record with its abstract, author keywords, ISSN and UCI; follow the works a given article cites; read journal-level citation metrics. OAK reaches the institutional repositories — theses, monographs, research reports, 고서 holdings and open-access articles contributed by member institutions. Half the tools need no credentials at all, so Korean material is reachable the moment the server is installed.

Records come back in the same response envelope the Japanese servers use, which is what makes genuinely trilateral work practical: Japanese, Korean and Anglophone scholarship on one question, read side by side in one format.

Related MCP server: Literatür MCP

What the receipts are for

A search you cannot re-run is a claim you cannot check. When a footnote rests on a database query, say that no article in this index uses a term before a certain year, the reader is asked to take the search on trust: which term, in which script, on what date, against which index and which version of it, and how far down the results the author went. Ordinary searching leaves none of that behind. This server leaves all of it. Every query-answering tool returns its envelope through the ledger, which appends one line to an append-only file: the term actually sent and its script, how the source matched it, how many records existed and how many came back, the diagnostics, the tool and its parameters, the server version, a timestamp, and the hash of the previous line. The hash makes the file a chain: a line cannot be altered, removed or reordered afterwards without the verifier saying so.

What that gives a researcher:

• A citable search. Name the receipt in the footnote (session slug, server, date, line hash) and a reader can see exactly what was asked and run it again against the same version.

• Negative findings that carry weight. "Not found" is evidence only if the search that produced it is on record, with its term, its script and its breadth.

• A method section that writes itself. korea-scholarship-mcp-ledger manifest <folder> summarises every query a project made, by server, script and session: the disclosure a journal, a data-availability statement or a research-integrity review asks for.

• A record of AI-mediated research. When a model chose the term, the receipt shows the term it chose and what came back, which is the thing to disclose about work done with an assistant.

• Nothing interpreted. The receipt is the source's own answer with credentials removed. The server does not summarise, rank or paraphrase, so the record is of the source, not of the tool.

Receipts are off until you name a folder (MCP_RECEIPT_DIR); each server then writes its own <server>.jsonl inside it, and MCP_RECEIPT_SESSION stamps a project or article slug on every line so one folder can serve several projects. korea-scholarship-mcp-ledger verify-dir <folder> checks the chains. The mechanics, the variables and what the envelope says when nothing is deposited are in the receipts section below.

Tools

Five of the nine work with no credentials at all — everything OAI-PMH, dbpia_links, plus status.

What the sources actually are

KCI indexes articles in Korean-registered scholarly journals. It does not index monographs, chapters, or dissertations. Its REST interface is a genuine query interface; its OAI-PMH interface is not.

OAK aggregates Korean institutional repositories — research reports, theses, monographs, 고서 holdings, OA articles — contributed unevenly by member institutions.

Both were probed live on 19 August 2026, and three properties shape how the tools are written:

1. OAI datestamps are ingest dates, not publication dates. A May 2019 harvest window returns articles published between 2010 and 2015. The often-repeated claim that KCI's OAI feed only exposes recent material is a misreading of this: the feed covers the corpus, it simply has no way to be asked anything. kci_harvest therefore filters client-side and says so in a diagnostic on every call.

1a. KCI's oai_dc is fully typed, and this server reads the types. Measured over 500 live records: identifier[type=artiId|uci|doi|citedCnt|regularity|journalInfo], an issn= attribute on 500/500, and lang="original|english" on every title and description. Version 0.2.0 asserted the opposite — "a positional, untyped bag" to be matched by pattern — and consequently discarded every ISSN, every abstract, and 371 real DOIs per 500 records. Pattern matching survives only as a fallback for identifiers that arrive untagged. Note that KCI also emits type="doi" elements containing nothing but the resolver prefix; those are normalised to null rather than passed through as identifiers.

2. OAK sends no resumptionToken. It declares noSetHierarchy, honours from/until, and caps a window at roughly 99 records with no continuation. A harvester that trusts the protocol will silently present a truncated window as a complete one. oak_harvest raises OAI_WINDOW_TRUNCATED when it hits the cap and tells you to slice the window.

3. OAK is not standard Dublin Core. It emits dc:title_h, dc:abstract_e, dc:publish_date, dc:location_org, dc:deep_link, dc:contents_url, and puts the material type in dc:keyword. Field presence varies by contributing repository. Unrecognised fields are preserved under extra.raw_fields rather than dropped.

Two further asymmetries are reported rather than smoothed over:

• KCI's articleSearch accepts keyword as a search field but omits author keywords, ISSN and UCI from its response. An empty keywords list is an artefact of the endpoint. kci_search says so on every call; kci_article recovers them.

• KCI answers HTTP 200 on failure, putting the error in outputData/result/resultMsg. A client that checks status codes reports an unregistered key as a successful empty search.

The response envelope

Every tool returns the envelope built by mediation.py and defined in response-schema.json, schema version 2.3.0 — typed query/script, matching_mode, graduated breadth, per-item matched_in, typed diagnostics, a loggable receipt, and attribution. Nothing is summarised or scored for you. kci_search also carries searched_for, the term actually sent with its detected script; the fetches and the harvests omit it, having chosen no term.

mediation.py 2.3.0 adds deposit reporting to 2.2.0, which was itself the reconciliation of a fork. Until 19 Aug 2026 two different files both called themselves 2.1.0: the Japanese copy had emit() — ledger persistence — but classified Hangul as latin; the Korean copy knew Hangul and the CJK extensions but had no emit(), so Korean queries never reached the deposit every Japanese query entered. 2.2.0 carries both, and is vendored byte-identical across cinii-mcp, jstage-mcp, ndl-mcp and this server. Everything in it is additive, so the Japanese servers adopt it without migration.

• detect_script() recognises Hangul and CJK Extensions B–G plus the Compatibility Supplement.

• title and source carry a ko slot alongside ja.

• emit() deposits the envelope to the hash-chained query ledger; ledger_available() reports whether it can, rather than leaving a silent no-op. As of v0.4.1 every query-answering tool in this server returns through emit(), rejections included — a query issued and refused was still issued — so Korean queries now enter the same deposit every Japanese query enters. korea_sources_status is the one exception: it chooses no term and answers no corpus, so it serialises with dumps() and instead reports the deposit state. Note the second gate: the ledger writes nothing unless MCP_RECEIPT_DIR (a receipts folder, one hash-chained file per server) or the legacy MCP_RECEIPT_LOG is set, and korea_sources_status now says which of the two gates is closed when nothing is being written.

title.romanized stays null unless the source supplies a romanisation. Neither KCI nor OAK does, and this server will not generate one: Revised Romanisation of a Korean name requires knowing the name, and a machine-transliterated string presented as bibliographic data is a fabrication with the shape of a fact.

Diagnostic codes

OK · NO_KEY · KCI_REJECTED · KCI_KEYWORDS_ABSENT · ZERO_CONJUNCTION · TRUNCATED · PAGE_PAST_END · REFERENCE_DEPOSIT_UNEVEN · BIBLIOMETRIC_SCOPE · SCRIPT_LATIN_QUERY · INGEST_DATE_NOT_PUBLICATION_DATE · CLIENT_SIDE_FILTER · OAI_MORE_AVAILABLE · OAI_INCOMPLETE · OAI_STALLED · OAI_PAGE_CAP · OAI_NO_RECORDS · OAI_ERROR · OAI_WINDOW_TRUNCATED · OAK_NONSTANDARD_DC · WINDOW_DOMINATED_BY_ONE_REPOSITORY · REDIRECTED · TRANSPORT_ERROR · API_ERROR · PARSE_ERROR · LINK_ONLY · NO_TERMS · DBPIA_NODE_UNRECOGNISED · RECEIPT_NOT_DEPOSITED · RECEIPT_WRITE_FAILED

Prerequisites

• Python 3.10 to 3.14 on PATH.

• Optionally, a KCI API key — free, self-registered, required only for the four REST tools.

Getting a KCI key

1. Register at open.kci.go.kr and apply for an Open API key.

2. The same key serves all five apiCode values (articleSearch, articleDetail, referenceSearch, citation, citationDetail).

KCI is also mirrored as four datasets on data.go.kr under 한국연구재단; that route issues a different key and is not used here.

Install

Three routes. All three give you the same server; pick by how much you want to see of it.

Python. The pip and source routes need Python 3.10 to 3.14; 3.10, 3.11, 3.12, 3.13 and 3.14 are tested in CI on Windows, macOS and Linux. The Claude Desktop bundle uses whichever of these is already installed, and has uv download one only if none is.

Getting Python

Every route needs Python 3.10 to 3.14. The Claude Desktop bundle uses one already on the machine and has uv download one only if none is; the other routes also need the venv module, which the official installers include.

• Windows. Download the 64-bit installer from python.org/downloads and run it; tick "Add python.exe to PATH" on the first screen. Afterwards py --version (the launcher the installer adds) or python --version in a new terminal should print 3.1x. If typing python opens the Microsoft Store instead, Windows has no Python yet: that Store page is a stub, and it is also what "'python' is not recognized" usually means.

• macOS. The python.org installer, or brew install python@3.13 with Homebrew. The /usr/bin/python3 that Xcode's command-line tools provide may be older than 3.10; python3 --version says.

• Linux. Your distribution's package: sudo apt install python3 python3-venv on Debian and Ubuntu, sudo dnf install python3 on Fedora. Or let uv provide one (next line).

• Any platform, with uv. uv installs Python itself: uv python install 3.13, then uv venv or the uvx route below.

One click: the Claude Desktop bundle

Download korea-scholarship-mcp-0.7.0.mcpb from the latest release and open it; Claude Desktop installs it. One bundle serves Windows, macOS (Apple Silicon and Intel) and Linux. Claude Desktop asks for KCI API key and a receipts folder at install time; the key is stored in the OS keychain.

The bundle carries the server's source and a lock file, nothing compiled, and needs no Python of its own. Claude Desktop builds the bundle's environment when you install it, with uv, a copy already on your PATH if there is one and otherwise one the app downloads for itself: uv takes a Python 3.10 to 3.14 already on the machine, downloads one only if there is none, and installs the locked libraries, roughly 40 MB, behind the install progress bar. Every launch then reuses that environment and takes under a second. Bundles 0.6.0 and 0.6.1 kept pyproject.toml one folder down, which made Claude Desktop skip that install step and download everything during the first connection attempt instead; on a slow or filtered network that attempt never completed and the app reported it could not connect to the extension server. Bundles before 0.6.0 vendored libraries compiled for CPython 3.12 only and failed on every other interpreter. See Troubleshooting.

From GitHub, pinned to a release

pip install "git+https://github.com/ckgerteis/korea-scholarship-mcp@v0.7.0"
# or, without an environment of your own:
uvx --from "git+https://github.com/ckgerteis/korea-scholarship-mcp@v0.7.0" korea-scholarship-mcp

installs the korea-scholarship-mcp console script and korea-scholarship-mcp-ledger. The tag is the thing to cite; @main gets whatever is current. Then register it in Claude Desktop (below), or let install.py do that.

The whole family

pip install "git+https://github.com/ckgerteis/bibliograph-mcp@v2.0.0" && bibliograph install

installs bibliograph-mcp, one server that carries this server's KCI and Open Access Korea tools, under the same names, alongside the rest of the family's (CiNii, J-STAGE, NDL, KCI and Open Access Korea, OpenAlex, Semantic Scholar) and sources that have no repository of their own (NDL Digital Collections full text, WARP, Crossref, Japanese name resolution, open-access PDF fetching), with one receipts folder and each credential asked for once. Since bibliograph 2.0.0 its sources are maintained in that package and may be ahead of this repository; bibliograph_status names the version of each. Register this server or bibliograph, not both, or Claude Desktop lists these tools twice. From a checkout of this repository, python install.py installs this server alone and python install.py --all the six separate servers, on Windows, macOS and Linux; install.ps1 remains for Windows.

From source

# from a clone
pip install .

# from a built wheel, whatever its version
pip install dist/korea_scholarship_mcp-*.whl

# from a clone, for development
pip install -e ".[dev]"

# without installing anything, straight from the repository
uvx --from "git+https://github.com/ckgerteis/korea-scholarship-mcp" korea-scholarship-mcp

Installing puts a korea-scholarship-mcp command on PATH. python -m korea_scholarship_mcp is equivalent.

The package is namespaced, so it shares an environment with cinii-mcp, jstage-mcp, ndl-mcp, openalex-mcp and semantic-scholar-mcp without colliding. Verify the install with:

python -c "import korea_scholarship_mcp as k; print(k.__version__)"

Do not use korea-scholarship-mcp --help as the check: unknown arguments are ignored, the server starts, reads end-of-input and exits 0, so it reports success whatever the state of the code.

Installing more than this one

Six independent packages. None imports another, none depends on another, and each installs and answers on its own — pip install . in this directory is a complete install of this server and nothing else.

They do share three things: a response envelope, a query ledger, and — if you run more than one — a receipts folder. install.ps1 is vendored byte-identical into all six and handles that on Windows; install.py is its cross-platform port. Both install this server by default, because cloning one repository is not a request for five more.

.\install.ps1                        # this server
.\install.ps1 -All                   # all six
.\install.ps1 -Servers korea_scholarship,cinii# a chosen subset

Nothing about where things go is decided for you. The script asks where to install (the virtual environment Claude Desktop will be pointed at), which folder receives the receipts, and which session slug to stamp on them, offering a neutral suggestion for each that Enter accepts; run without a terminal it does not guess, and stops unless --venv and --receipts-dir (or --no-receipts; -VenvDir and -ReceiptsDir for install.ps1) say so. Whatever subset you name is registered against one receipts folder, asked for once. The script prefers a sibling checkout to the network, carries across credentials already registered rather than asking again, leaves servers it was not asked about alone, and stops rather than guessing where the servers already registered disagree about the folder or the session slug. It also asserts that ledger.py and mediation.py are byte-identical across everything it installed, so two envelope versions cannot end up in one environment unnoticed.

Any other MCP client

Nothing here is specific to Claude. The server speaks the Model Context Protocol over stdio and nothing else: any client that can start a process and talk JSON-RPC to it (Claude Code, Cursor, VS Code and Continue, Zed, LibreChat, a script of your own using an MCP SDK) can use it. The Claude Desktop bundle and the installers are conveniences for one client; the server underneath is the same console script. Register it anywhere by giving the client the absolute path of the console script and, optionally, the environment:

{
  "mcpServers": {
    "korea_scholarship": {
      "command": "/absolute/path/to/.venv/bin/korea-scholarship-mcp",
      "env": {
        "KCI_API_KEY": "your key (optional; four tools need none)",
        "MCP_RECEIPT_DIR": "/absolute/path/to/receipts",
        "MCP_RECEIPT_SESSION": "project-or-article-slug"
      }
    }
  }
}

Claude Code takes the same thing on the command line:

claude mcp add korea_scholarship -- /absolute/path/to/.venv/bin/korea-scholarship-mcp

On Windows the path ends in \.venv\Scripts\korea-scholarship-mcp.exe. MCP_RECEIPT_DIR and MCP_RECEIPT_SESSION are optional; without them the server runs and every envelope says RECEIPT_NOT_DEPOSITED. The stdio transport is the only one: there is no HTTP endpoint to expose, and nothing to host.

Troubleshooting

"Server disconnected" is all Claude Desktop says when the server process exited before or during the handshake, whatever the reason. The reason is in the log:

• Windows: %LOCALAPPDATA%\Claude\Logs\mcp-server-<name>.log (builds before August 2026: %APPDATA%\Claude\logs) (the extension's display name, or the key under mcpServers), with mcp.log beside it for the app's side of the conversation.

• macOS: ~/Library/Logs/Claude/mcp-server-<name>.log and mcp.log.

• Linux: ~/.config/Claude/logs/.

Read the last launch from the bottom up. Three shapes account for nearly every report:

• A Python traceback ending in ImportError or ModuleNotFoundError (for example No module named 'pydantic_core._pydantic_core'). The interpreter started, the code was found, and a compiled library did not match that interpreter. This is what every bundle before 0.6.0 did on any Python other than 3.12. Install the current bundle, or use the pip route, which resolves wheels for the interpreter you install into.

• 'python' is not recognized, spawn python ENOENT, or a line from the Microsoft Store: no interpreter was found on the PATH Claude Desktop constructs. Nothing of this server ran. The current bundle does not launch python at all; for the pip route, register the console script by absolute path as shown above.

• A line from uv (error: ..., or a download that never finished), or Unable to connect to extension server with nothing from the server in the log: the environment was not built. From 0.6.2 the app builds it at install time; look in main.log beside the server log for lines tagged [UV Runtime], which record the download and the uv sync, and for missing pyproject.toml, which means a bundle older than 0.6.2. If the install-time build failed (no network, a proxy that blocks github.com or pypi.org), reinstall the bundle once the network is back; the launch also rebuilds the environment itself, so a second launch on a working network completes.

The bundle's own entry point writes one line naming the interpreter, its path and the supported range before re-raising an import failure, so a log from 0.6.0 onwards says which of these it is.

Configuration

cp .env.example .env

KCI_API_KEY=your_kci_api_key_here

Claude Desktop

If the package is installed, point at the console script:

{
  "mcpServers": {
    "korea-scholarship": {
      "command": "C:\\path\\to\\.venv\\Scripts\\korea-scholarship-mcp.exe",
      "env": {
        "KCI_API_KEY": "your_kci_api_key_here"
      }
    }
  }
}

Or run it from a clone without installing:

{
  "mcpServers": {
    "korea-scholarship": {
      "command": "C:\\path\\to\\.venv\\Scripts\\python.exe",
      "args": ["-m", "korea_scholarship_mcp"],
      "env": {
        "KCI_API_KEY": "your_kci_api_key_here"
      }
    }
  }
}

Omit the env block entirely to run the four keyless tools.

A note on the MCP SDK

mcp 2.0.0 removed mcp.server.fastmcp. This server imports FastMCP where it exists and falls back to MCPServer where it does not, so it runs on either. The same shim was applied to cinii-mcp and jstage-mcp on 19 August 2026; before that, both imported mcp.server.fastmcp directly while pinning mcp[cli]>=1.2.0 with no upper bound, so a fresh install of either resolved to 2.0.0 and failed at import.

Credential handling

The KCI key travels in the query string, which makes it leak-prone in two specific ways this server closes:

• httpx logs every request URL at INFO. _silence_http_logging() mutes it and strips any stdout handler — necessary anyway, since stdout carries JSON-RPC.

• Transport and status exceptions embed the request URL. Every message bound for the client passes through _redact(), and the receipt is built from parameters with credentials removed rather than masked.

Tests

python -m pytest tests -q                # offline, against fixtures captured 19 Aug 2026
RUN_LIVE=1 python -m pytest tests -q     # also exercises the live KCI endpoints
RUN_LIVE_OAK=1 python -m pytest tests -q # adds OAK; needs a network that reaches oak.go.kr

The live tests guard the claims this README rests on: that a KCI ingest window returns older publications, that KCI's identifiers are typed, that max_records is a cap rather than a hint, and that a resumption harvest does not record a date window it never sent. The OAK test is gated separately and fails loudly if OAK is unreachable rather than passing on an unexercised branch.

tests/smoke_stdio.py starts the installed console script over stdio, performs the MCP handshake, and checks tools/list against the tool table above; RUN_LIVE=1 … <tool> '<json params>' adds one live call.

Known limits

The four KCI REST tools have never seen a live response — there is no API key. Their field mapping follows the published documentation and is unverified against the wire; the success/failure test is deliberately structural (records present means success) so that neither a chatty success message nor a terse rejection is misread. Treat REST output as provisional until a key exists.

What this server does not cover

ScienceON (KISTI) — deliberately out of scope. Its gateway requires an AES-256-CBC token built from a registered MAC address, plus a registered public IP. rubato103/scienceon-mcp already implements it against live credentials and is hardened against the exact credential-leak path described above; install it alongside rather than duplicating untestable auth code:

claude mcp add scienceon -- uvx --from "git+https://github.com/rubato103/scienceon-mcp" scienceon-mcp

RISS (KERIS) — the search API exists at https://www.riss.kr/openApi and covers theses, domestic and foreign articles, monographs, research reports and serials, but keys are issued only to Korean non-profit institutions and universities, each application approved by KERIS staff; individuals cannot apply. Whether a non-Korean university qualifies is untested. If a key is ever obtained, RISS belongs in this server.

DBpia (Nurimedia) — keys are open and generous (2,500 calls a day), but the terms of use restrict the service to non-commercial purposes and forbid copying, storing or transmitting search results, which are to be displayed in real time and unaltered. That is incompatible with harvesting into a reference manager, a corpus index, or a register. The constraint is the licence, not the API. dbpia_links stays inside it by not calling the API at all: it builds the address of DBpia's own search or article page and returns it, so the user sees DBpia's answer in a browser and nothing of it passes through this server or its receipts.

korea_sources_status reports all three of these in situ, so the omission is visible from inside the tool rather than only in this file.

Usage rules

• KCI and OAK are public-sector services with no published rate limit. Harvest considerately; slice windows rather than hammering wide ranges.

• Metadata retrieved here is bibliographic. Full text sits behind whatever terms the holding repository sets — OAK's contents_url points into member repositories, each with its own licence.

• Attribution strings are returned in every envelope; carry them into anything published.

Citation

If this software supports your research, please cite it. See CITATION.cff, or use the "Cite this repository" button on GitHub.

License

MIT © 2026 Christopher Gerteis.

This license covers the server code only. It grants no rights over KCI or OAK data, which remain governed by the terms of the National Research Foundation of Korea and the National Library of Korea respectively.

Disclaimer

A research tool, maintained on a best-effort basis and provided "as is", without warranty. Not affiliated with or endorsed by the National Research Foundation of Korea, the National Library of Korea, KERIS, KISTI, or Nurimedia.

Author

Dr Christopher Gerteis, SOAS University of London.