Skip to content

DR-NTU

Service overview · CLI reference · MCP reference

Search the public DR-NTU research repository, browse collections and inspect publication and dataset metadata. Try ntu research search "machine learning".

Service notes

Public publication, dataset, and OAI-PMH metadata lookups in the shared all-abt-ntu-mcp server. No credentials, keyring, Chromium, or extra dependencies are needed. See the getting started guide for installation and MCP client configuration.

All results have retrieved_at, source_url, and cached. Responses are cached in memory for five minutes, with at most 64 entries and 16 MB per REST/OAI client. refresh: true refetches the requested response. Each client pools HTTPS connections, permits up to two in-flight reads with 250 ms default request pacing, limits responses to 8 MB, and uses a 20-second timeout. Requests to researchdata.ntu.edu.sg are spaced at least 60 seconds apart per process. It retries transient failures once and honors short Retry-After delays; longer server-requested delays return an error rather than retrying early. Public HTTP settings are documented in the shared runtime guide.

Access policy review

Reviewed 6 November 2026: DR-NTU (Data) general terms require scraper bots to follow robots.txt and prohibit excessive load; the API terms require an identifiable application and prohibit unreasonable bandwidth and impairment. The public client identifies itself as all-abt-ntu-mcp/0.1 (public NTU lookup) and does not retrieve dataset file bodies. The research-data robots policy specifies a 60-second crawl delay and disallows crawling paths other than listed exceptions, including the dataset API. The dataset tool is intended for individual API lookups, not crawling or bulk harvesting; the conservative 60-second pacing also applies to those API requests. The publication repository policy restricts specific UI paths and /server/opensearch; the REST and OAI paths used here were not listed as disallowed at review time. Recheck both policies and applicable terms before release or any broader usage. These process-local limits do not coordinate multiple replicas.

DSpace listings use zero-based page and page_size from 1–50. Follow pagination.next_page until null. Total counts remain null when the source omits them, as happens with individual search facets. Malformed records or inconsistent result pages fail rather than silently appearing complete.

Sources

Endpoint Purpose
https://dr.ntu.edu.sg/server/api/discover/search/objects Paginated publication/profile search, scopes, and facet filters
https://dr.ntu.edu.sg/server/api/core/items/{uuid} Published item metadata, abstracts, and rights
https://dr.ntu.edu.sg/server/api/core/communities Community browsing
https://dr.ntu.edu.sg/server/api/core/collections Collection browsing
https://dr.ntu.edu.sg/server/api/core/communities/{uuid}/subcommunities Community hierarchy
https://dr.ntu.edu.sg/server/api/core/communities/{uuid}/collections Collections within a community
https://dr.ntu.edu.sg/server/api/discover/facets Filter definitions
https://dr.ntu.edu.sg/server/api/discover/facets/{name} Filter values, counts, and exact query filters
https://dr.ntu.edu.sg/server/api/core/items/{uuid}/bundles Item file-group discovery
https://dr.ntu.edu.sg/server/api/core/bundles/{uuid}/bitstreams File metadata and content links
https://researchdata.ntu.edu.sg/api/datasets/:persistentId Latest dataset metadata with persistentId=doi:...
https://researchdata.ntu.edu.sg/api/datasets/:persistentId/versions/{version} Explicit published dataset release
https://dr.ntu.edu.sg/server/oai/request OAI-PMH Identify, ListMetadataFormats, ListSets, ListRecords, ListIdentifiers, and GetRecord

The API root was checked to identify the repository's DSpace/CRIS format. Bundle details and individual bitstream metadata repeat fields already included in listings; separate format lookups were inspected but are omitted to avoid a request for every file. OAI-PMH supports incremental metadata retrieval, while the REST search tools provide keyword and facet searches.

Primary references: DSpace REST contract, search contract, Dataverse Native API, and OAI-PMH 2.0. Current NTU response shapes were also checked directly over HTTPS.

Tests

Offline parsers, pagination, filters, caching, network failures, CAPTCHA/error handling, and MCP schemas use synthetic records:

uv run python -m unittest discover -s tests/drntu -t . -v

Opt-in live stdio checks search pages, publication filtering, item abstracts, browsing, facets and their filters, file discovery, and both dataset version endpoints. Prints statuses and never downloads research file bodies:

uv run python -m tests.drntu.live_smoke

OAI-only live stdio checks exercise all six protocol verbs, provider continuation, date filters, alternate-format XML, empty results, and expired/invalid token errors. Prints statuses and counts without saving or printing research metadata:

uv run python -m tests.drntu.live_oai_smoke

Limitation

Only publicly exposed metadata is supported. No file downloads, full-text extraction, authenticated repository access, dataset search, or automatic repository-wide OAI harvesting is implemented. OAI timestamps describe repository changes, not publication dates. Continuation tokens can expire and NTU controls source-page sizes. Metadata may describe restricted or embargoed content. Dataset access can be interrupted by CAPTCHA; availability during one smoke test is not a guarantee of continued automated access.