Access guide for programs and AI agents
This database serves readers, search-engine crawlers, AI agents and your own scripts alike. Everyone gets the same content under the same access rules; only the entrance differs: readers use the pages, programs can fetch JSON directly.
Data interface
Read-only, JSON responses, no registration and no key. The machine-readable description is at:
https://ejournal.press.ntu.edu.tw/index.php?act=Opendata
https://ejournal.press.ntu.edu.tw/index.php?act=Opendata/journals- All journals: ID, title, discipline, access level, article count, latest issue
https://ejournal.press.ntu.edu.tw/index.php?act=Opendata/journal/J17- One journal: basic details and all of its issues
https://ejournal.press.ntu.edu.tw/index.php?act=Opendata/issue/J17_0_104- Table of contents of one issue
https://ejournal.press.ntu.edu.tw/index.php?act=Opendata/article/1- One article: full citation, abstract, keywords and full-text status
https://ejournal.press.ntu.edu.tw/index.php?act=Opendata/search&q=…- Search, with the same parameters as the search page
When searching, run the search once in Chinese and once in English: most records have both Chinese and English titles and abstracts, and the two languages do not return exactly the same results. Each search result carries the beginning of the abstract and the keywords; years can be given as a range (year_from, year_to) of at most ten years.
In its head section, every page of the site carries <link rel="service-desc"> pointing to the machine-readable description above.
MCP endpoint for AI assistants
AI assistants that support the Model Context Protocol (MCP) can add this database as a set of tools. In the assistant’s MCP settings, add a remote (HTTP) server with this URL:
https://ejournal.press.ntu.edu.tw/index.php?act=Opendata/mcp
No account or key is needed. Once added, the assistant sees the five read-only tools below. They offer the same functions as the URLs above, under the same access rules.
search_articles- Search articles
get_article- One article: full citation, abstract, keywords and full-text status
list_journals- All journals: ID, title, discipline, access level, article count, latest issue
get_journal- One journal: basic details and all of its issues
get_issue- Table of contents of one issue
Full text is not sent through MCP: the tools return its status and URL, and the assistant reads it under the site’s access rules. A cloud AI assistant connects from its vendor’s servers, which are off campus, so it cannot read campus-only full text; an assistant running on a campus computer can.
Access to full text
There is no separate full-text endpoint. Each article response includes full_text: the status of the full text, whether this connection may read it, and its URL. That is the same URL the site itself uses, with the same access check.
- Full text of open journals can be read from anywhere.
- For journals marked “Campus only”, full text can be read only from the NTU campus network. The same applies to programs.
- Citation-only articles have no full text to read.
Usage limits
Usage by programs is counted separately from readers, so programs never use up the capacity reserved for readers. In turn, a program over its limit is only asked to try again later; it is not blocked.
Bulk crawlers and AI assistants acting for a user are also counted separately: a crawler that uses up its own allowance does not affect assistants. A crawler counts as a crawler whether it fetches pages or the data interface.
| Source | Per source, per minute | All sources combined, per minute |
|---|---|---|
| Data interface, and AI assistants fetching for a user (requests) | 120 | 300 |
| Crawlers, and other self-identified programs reading pages (requests) | 600 | 300 |
| Full text opened by AI assistants (files) | – | 10 |
| Full text opened by crawlers and programs (files) | – | 30 |
| Filtered search result pages read by programs (requests) | – | 20 |
Over the limit, the response is HTTP 429 with a Retry-After header. Wait that many seconds before trying again.
Filtered search result pages (selected, excluded, or by discipline) come in endless combinations and repeat what the article pages contain, so robots.txt closes them to robots, and such requests from programs share a very low limit. To query by conditions, use the search of the data interface. Requests that neither identify themselves nor show signs of a real browser count toward the same limit.
Every response from the data interface carries RateLimit-Policy and RateLimit headers: the limits that apply to this source, how many requests are left in the current minute, and how many seconds until the count restarts.
Fetch each full text in a single request: when a program fetches one file in several ranges, every range counts once. HEAD requests and unchanged (304) responses do not count.
Programs that do not identify themselves (a browser-like User-Agent) are treated as readers. The limit per source, per minute, is 120 requests; going over is refused temporarily, and sources that keep doing so are blocked. Please put the name of your program and a contact in the User-Agent.
When nothing has changed
Every page and every data-interface URL supports ETag and Last-Modified. Send If-None-Match or If-Modified-Since; unchanged content returns 304 and need not be downloaded again. In data-interface responses, meta.updated is the time the data last changed.
Conditions of use
- Full texts are for personal research and teaching only.
- Journals marked “Campus only” can be read in full only on the NTU campus network.
- Do not download in bulk with automated tools.
For questions, or if your research needs higher limits, contact National Taiwan University Press:ntuprs@ntu.edu.tw
Asking an AI assistant to use this database
If the assistant supports MCP, adding the MCP endpoint as described above is the easiest way. If not, give it the URL below. Everything it needs to know is described there:
https://ejournal.press.ntu.edu.tw/index.php?act=Opendata