# Hummingtomidi - expanded reference for agents

## What the site is

Hummingtomidi is a small static website of browser-based audio conversion tools.
It is hosted on Cloudflare Pages. There is no backend, no database, no user
accounts, no upload endpoint, no advertising, no analytics and no third-party
scripts. Files a visitor selects never leave the visitor's device.

## Public pages

- `/` — MP3 to MIDI Converter (audio to MIDI)
- `/wav-to-midi` — WAV to MIDI Converter (wav to midi)
- `/audio-to-midi` — Audio to MIDI Converter (audio to midi)
- `/midi-to-wav` — MIDI to WAV (midi to wav)
- `/about` — About (about)
- `/privacy-policy` — Privacy Policy (privacy)
- `/contact` — Contact (contact)
- `/ai/` — AI & Agent Discovery (agent discovery)

## How conversion works (audio to MIDI)

The browser decodes the selected file with the Web Audio API, then a monophonic
pitch tracker (autocorrelation) produces note events (pitch, start, duration).
A Standard MIDI File (SMF) is written in JavaScript and offered as a download.
It is monophonic and works best on one clear melodic line. Inputs accepted:
MP3, WAV, OGG, M4A, AAC, FLAC, OPUS. Long recordings are capped at the first
three minutes. One file at a time. No accuracy figure is claimed.

## How rendering works (MIDI to WAV)

A .mid / .midi file is parsed in the browser (SMF header, tracks, tempo map,
note-on/note-off pairing). The notes are synthesised with OfflineAudioContext
and encoded as 16-bit PCM RIFF/WAVE, then played and offered as a download.
A MIDI file contains note instructions, not sound; timbre and exact duration
depend on the synthesiser, which is why different renderers sound different.

## Machine-readable entry points

- robots.txt: https://hummingtomidi.com/robots.txt (AI crawler rules, Content-Signal, Agentmap, Sitemap)
- sitemap: https://hummingtomidi.com/sitemap.xml
- API catalog (RFC 9727 linkset): https://hummingtomidi.com/.well-known/api-catalog
- OpenAPI description: https://hummingtomidi.com/ai/openapi.json
- API documentation: https://hummingtomidi.com/ai/docs
- MCP server card: https://hummingtomidi.com/.well-known/mcp/server-card.json
- Agent Skills index: https://hummingtomidi.com/.well-known/agent-skills/index.json
- Skill: https://hummingtomidi.com/ai/skills/site-lookup/SKILL.md
- ARD catalog: https://hummingtomidi.com/.well-known/ai-catalog.json
- DNS-AID index: https://hummingtomidi.com/ai/index.ilang

## Authentication status

This is a public, unauthenticated static site. Authentication is NOT available.
The OAuth discovery documents describe a planned, under-construction contract
only. No accounts exist, no tokens are issued, no identity data is collected.

## Privacy

No analytics, no advertising, no cookies, no third-party scripts. See
https://hummingtomidi.com/privacy-policy. If advertising or analytics are added in the future the
policy will be updated and any cookie use disclosed.


## Live endpoint contract

The lookup operation is served by a real Cloudflare Pages Function, not a static stub. GET /api/site-lookup?q=... reads the same published /ai/site-lookup.json data and returns {query, count, results}. Known item "midi to wav" returns the canonical /midi-to-wav URL; an unknown term returns count: 0. POST /mcp implements JSON-RPC initialize, tools/list and tools/call for the single site-lookup tool. Both are read-only and require no authentication: the site has no identity store. Planned authorization endpoints deliberately return 503 and never read or store identity data.