jp-lit-mcp

An MCP server for Japanese literature research designed to allow AI agents to perform cross-database searches across NDL Search, NDL Digital Collections, CiNii Research, J-STAGE, IRDB, JDCat, nihuBridge, and the Minutes of the National Diet and Imperial Diet.

MCP provides the tools for searching and retrieval, while the included Skills assist in dialogue with "which DB to use," "what search terms to try," and "how to evaluate the results."

First, choose your application

The most common hurdle in setup is the difference in MCP/Skills configuration for each application. Please start by opening the instructions for the application you are using.

As a prerequisite, Node.js 18 or higher and npm are required. For standard use, you do not need to clone the repository; you can execute the MCP server and Skills installer via npx.

• Installation instructions for Cursor

• Installation instructions for Claude Code

• Installation instructions for Codex CLI

• Installation instructions for Codex App

• Installing Skills with GitHub CLI

Related MCP server: rust-research-mcp

Quick Start

1. Register npx -y jp-lit-mcp as an MCP according to your application's installation instructions.

2. Install Skills using npx -y jp-lit-mcp install-skills <app>.

3. Request research on the application.

For troubleshooting after installation, you can use the lightweight diagnostic command.

npx -y jp-lit-mcp doctor

This command checks for Node.js 18 or higher, package versions, included Skills, write permissions for cache/exports directories, and the presence of CINII_RESEARCH_APP_ID. CINII_RESEARCH_APP_ID is used for CiNii-related and KAKEN API tools. It does not perform live API access to external databases.

Only if you wish to develop or add sources, clone this repository and run npm install / npm run build / npm run smoke:mcp.

Example of an initial request:

文献DBで、近代日本の労働文化について、論文と図書を探してください。

文献DBを始めます。明治期の俳句雑誌について、最初に見るべき資料と、使うべき DB を教えてください。

文献検証で、この文章に出てくる文献の実在性を確認してください。

What you can do

• Search for books, papers, magazine articles, meeting minutes, and research data.

• Check holdings and bibliographic details via NDL / CiNii Books, etc.

• Refer to research manuals and case studies from the Collaborative Reference Database.

• Handle OCR full text, page coordinates, and illustrations/figures from NDL Digital Collections.

• Verify the existence of literature mentioned in pasted text.

• Filter, integrate, and reorganize saved research results, and export them in Markdown / JSON / CSL JSON.

Research results exported in CSL JSON can be passed to reference management and citation processing tools such as Zotero, Pandoc, and citeproc-based tools.

For details on supported sources and MCP tools, refer to the Technical Reference.

Why use Skills?

While you can search with MCP alone, the user must think about source selection, search term expansion, and result evaluation.

By using the jp-lit-research Skill, you can create a research plan before searching, and assemble sources and search terms while referring to the Collaborative Reference Database or NDL Research Navi as needed. Research is not intended to end with a single search; it is based on the premise of re-selecting the next query or DB while looking at candidates.

When returning results, we include not only bibliographic information but also highlights from full-text searches and short excerpts from summaries or tables of contents whenever possible, in a format that explains "why this material was presented." Page location identification is performed with a separate tool only when necessary.

jp-lit-mcp is not intended for LLMs to make final decisions on understanding the content or academic positioning of literature. Even for literature where the main text has not been read, it may tentatively organize it based on titles, abstracts, tables of contents, book reviews, publisher introductions, and fragments on the Web. In such cases, it will explicitly state that it is not a reading of the main text and what the basis for the organization is. online=true or links to PDF / HTML / Digital Collections only indicate that there is an entry point online and do not mean that the agent has read the main text.

Candidates are not just listed flatly; they are tentatively prioritized for research confirmation based on material type, publisher/media, author attributes, citation/review status, and main text confirmation status. However, the value of literature is not determined solely by the publisher or media. Confirmation priority is a working guideline for humans to decide what to look at next.

It also supports organizing results after searching. Operations such as sorting current results, narrowing down to online publications only, or taking differences/commonalities with previous results are, in principle, performed by reusing saved results. You can cross-search and integrate not only current search results but also past saved search results. cache.hit=true in jp_lit_search indicates that the cache from which it was reused was used. Old local caches can be deleted after checking candidates with jp_lit_prune_cache.

Auxiliary tools using Web NDL Authorities can also be used to confirm pen names, aliases, and subject headings of personal names. For example, you can confirm the relationship between names like Takehiro Irokawa and Tetsuya Asada, and decide whether to search by name or collectively. You can also reverse-lookup subject headings from classifications like NDC to create search term candidates for finding unknown books.

For standard Skill installation, we recommend npx -y jp-lit-mcp install-skills <app> found in the install guide for each app. There is another route using gh skill install from the GitHub CLI, but this is for advanced users. For details, refer to Installing Skills with GitHub CLI.

The jp-lit-verification Skill extracts Japanese literature candidates that appear in other services' answers or your own text, and confirms them by categorizing them into confirmed existence / partial match / suspected non-existence / suspected confusion.

For detailed usage, refer to the Usage Guide.

Main Supported Sources

Commonly used sources are as follows:

• ndl_catalog: Entry point for checking the National Diet Library and holding information

• ndl_digital: NDL Digital Collections

• cinii_articles / cinii_books: Papers, university library books/magazines

• jstage_articles: Academic journals/research papers

• irdb: University institutional repositories

• nihu_bridge: Cross-search of humanities-specialized databases

• kokkai_minutes / teikoku_minutes: Minutes of the National Diet and Imperial Diet

• jdcat: Research data in humanities and social sciences

• japan_search: Cultural properties, museums, regional materials

A list and implementation notes are summarized in the Technical Reference.

Documentation

• Usage Guide: Actual request examples, research flow, how to read output

• Installing Skills with GitHub CLI: Another route using gh skill install

• Technical Reference: Sources, MCP tools, environment variables, constraints, development/verification commands

• Data Usage Conditions Memo: Display requirements and usage conditions for external DBs / APIs

• Project Status: Current state, recent updates, post-release notes

License

The code in this repository is under the MIT License. See LICENSE for details.

However, the data usage conditions for external DBs / APIs accessed by MCP are different. Points of caution change between research use on personal devices and operation of public services or shared servers that accumulate search results and provide them to multiple users. For conditions regarding redistribution, display, commercial use, and mirror-like storage, please check the Data Usage Conditions Memo and the terms of each provider.