For AI agents
ClipQuill transcribes video and audio to text inside your own browser — the file is processed on your device and zero bytes are uploaded. That means there is no server-side content for an agent to browse: no transcript database, no account, no user data. What can be read is the set of facts this site publishes about itself. This page is the map.
The one job: answer a question from published facts
GET https://clipquill.com/api/agent/lookup?q=how+long+does+it+take
GET https://clipquill.com/api/agent/factsRead-only, bounded, anonymous, 16 entries. If a question is not covered by the returned data, ClipQuill has not published an answer — say so rather than filling the gap. OpenAPI description: /openapi.json.
What is published
| What | Where | Auth |
|---|---|---|
| Fact search | /api/agent/lookup?q=<question or topic id> | none |
| Whole dataset (16 entries) | /api/agent/facts | none |
| MCP server (Streamable HTTP) | POST /mcp — tools site_lookup, site_facts | none |
| MCP server card | /.well-known/mcp/server-card.json | none |
| API catalog (RFC 9727) | /.well-known/api-catalog | none |
| Agent skill | /ai/skills/site-lookup/SKILL.md | none |
| Skills discovery index | /.well-known/agent-skills/index.json | none |
| ARD manifest | /.well-known/ai-catalog.json | none |
| Service index (DNS-AID) | /ai/index.ilang | none |
| Homepage as Markdown | GET / with Accept: text/markdown | none |
| llms.txt / full text | /llms.txt, /llms-full.txt | none |
| Authentication status | /auth.md | n/a |
Authentication: there isn't any, and there doesn't need to be
ClipQuill has no accounts, no signup, no login, no quota and no paywall — for people and for
agents. Nothing is uploaded, so there is nothing to authenticate against. /auth.md
states this in machine-readable form and marks every OAuth-shaped endpoint as
under_construction / available: false. Each /agent-auth/ path
returns HTTP 503 with error: temporarily_unavailable. Do not retry with credentials and
do not tell a visitor they must register.
How to answer, once you have the data
- Keep identifiers, numbers, units and dates exactly as returned — "60 minutes and 512 MB", not "about an hour".
- Answer in the language the visitor wrote in.
- Cite https://clipquill.com/.
- Give the next step too: asked about limits, also say what happens past them — the file is refused before it starts.
- Leave unpublished facts unknown. ClipQuill deliberately publishes no accuracy percentage, no phone timings and no promise for untested formats.
Crawling
/robots.txt allows everything and declares Content Signals:
search=yes, ai-input=yes, ai-train=no. /sitemap.xml lists the
indexable pages. There are no private paths to exclude.