Keio University Agentic Access

generated: '2026-09-01'
method: derived
authorship: API Evangelist
probe_basis: live probe sweep of Keio University surfaces, 2026-09-01
source: >-
  Direct probes on 2026-09-01 of the agent-facing conventions on Keio's main and service hosts.
  Every status code below is from a real fetch with a browser User-Agent.
provider: Keio University
providerId: keio
description: >-
  What an AI agent arriving at Keio University can actually do. The finding is mixed in an
  interesting way: the institution publishes none of the modern agent-facing conventions and, on
  its main web estate, not even the old ones — but the two surfaces it does operate are among the
  friendliest an agent could ask for, because they are anonymous, standards-described and
  self-describing. An agent that knows OAI-PMH or IIIF can consume Keio's scholarly and digitised
  holdings today with no key, no negotiation and no account.

conventions:
  - name: llms.txt
    url: https://www.keio.ac.jp/llms.txt
    status: 404
    present: false
    note: A real HTTP 404 returning the site's Japanese not-found page, not a soft-404.
  - name: robots.txt (main site)
    url: https://www.keio.ac.jp/robots.txt
    status: 404
    present: false
    note: >-
      The primary institutional web estate publishes no robots.txt at all. Recorded as observed;
      the absence means there is no crawl directive of any kind for the main site.
  - name: robots.txt (KOARA repository)
    url: https://koara.lib.keio.ac.jp/robots.txt
    status: 200
    present: true
    content: 'User-agent: * / Allow: /'
    note: Maximally permissive — the repository invites harvesting.
  - name: robots.txt (Digital Collections)
    url: https://dcollections.lib.keio.ac.jp/robots.txt
    status: 200
    present: true
    note: Drupal's stock robots.txt, unmodified.
  - name: security.txt
    url: https://www.keio.ac.jp/.well-known/security.txt
    status: 404
    present: false
  - name: MCP server
    present: false
    note: No Model Context Protocol server, manifest or endpoint found on any Keio host.
  - name: Agent card (/.well-known/agent.json)
    present: false
    note: >-
      Not found, and deliberately not derived. Agent cards are search-only in this pipeline; one is
      never written on a provider's behalf.
  - name: OpenAPI published by Keio
    present: false
    note: >-
      Keio publishes no OpenAPI, no AsyncAPI, no apis.json and no WADL. The two contracts in this
      repository were written by API Evangelist from probes and are marked as such in their
      x-provenance blocks.

agent_consumable_today:
  - surface: KOARA OAI-PMH
    baseURL: https://koara.lib.keio.ac.jp/xoonips/modules/xoonips/oai.php
    x-operator: institution
    why: >-
      Anonymous, no key, no rate limit encountered, responses self-describe against a published
      schema, and the full set hierarchy is enumerable in one call. An agent can harvest the entire
      scholarly record of the university without asking anyone.
    caveat: >-
      Protocol errors arrive with HTTP 200 and the failure inside the XML body. An agent branching
      on status code alone will silently mistake failures for successes.
  - surface: IIIF Presentation and Image APIs
    baseURL: https://dcollections.lib.keio.ac.jp/sites/default/files/iiif/
    x-operator: institution
    why: >-
      Anonymous JSON-LD, IIIF-conformant, with an image service that answers arbitrary region, size
      and rotation requests. An agent can retrieve and reason over 656 folios of the Gutenberg
      42-line Bible without an account.
    caveat: >-
      Archive codes are undocumented. Manifest URLs must be resolved from the site's HTML, which is
      the one real barrier to autonomous discovery on this surface.

agent_blocked:
  - surface: keio.figshare.com
    status: 202
    detail: >-
      Returns HTTP 202 with a zero-byte body to a scripted client — a bot interstitial. Recorded as
      a fact about the platform's edge, not about Keio.
  - surface: www.linkedin.com/school/keio-university/
    status: 999
    detail: LinkedIn's standard bot rejection. The pointer is live for a human; not a dead link.