ht-ml.app — HTML hosting with an API built for agents

ht-ml.app

4 min read Original article ↗

HTML hosting · API for agents

Go live straight from Claude Code or Codex in under a minute.

To get started, ask Claude Code or Codex to deploy your HTML file with https://ht-ml.app. Optimized for deploying prototypes, rich architecture diagrams, code reviews, decks, illustrations, and more straight from agentic environments. POST a single HTML document and get a public URL back instantly. No accounts, no dashboards, no API keys to provision.

No signup JSON in · JSON out WebMCP-ready

Pages are public by default — anyone with the link can view them, and they may be crawled or indexed. To keep something private, set a password when you create the site, so only people you share it with can see it. See the Terms.

Agent skill

Install the skill for your agent

Prefer a turnkey setup? The open html Agent Skill teaches your agent the whole workflow — create, update, assets, password protection — and bundles 20 ready-made page templates. One command installs it for Claude Code and Codex.

Install · Claude Code + Codex

curl -fsSL https://raw.githubusercontent.com/nsmith/html/main/install.sh | sh

Alternative · skills CLI — installs to all detected agents

npx skills add nsmith/html

Claude Code Codex OpenClaw Hermes + any Agent Skills client

Open format (agentskills.io) · source & per-client install paths on GitHub.

Quickstart

One request, one live site

Send your HTML to the API. The response hands you the public URL and the secret key you'll use for any later edits.

Request

curl -X POST https://api.ht-ml.app/v1/sites \
  -H "Content-Type: application/json" \
  -d '{"html_content": "<h1>Hello from my agent</h1>"}'

Response · 200 — values in {braces} are placeholders

{
  "site_id": "{site_id}",
  "url": "https://{site_id}.ht-ml.app/",
  "update_key": "{update_key}",
  "status": "active"
}

Read the live URL from the url field of your response — each site gets its own subdomain, https://{site_id}.ht-ml.app/, so relative asset paths just work (<img src="images/logo.png">). Keep the update_key secret — it's returned only once and grants write access.

The flow

How an agent uses ht-ml.app

  1. Create the site

    POST html_content to /v1/sites. The HTML is scanned for unsafe content, then stored. You get back a site_id and an update_key.

  2. Discover what's missing

    GET /v1/sites/{site_id} to enumerate the assets your HTML references (images, video) and which are still missing.

  3. Upload referenced assets

    For each missing asset, POST the file to /v1/sites/{site_id}/assets with your update_key as a Bearer token. Paths mirror the src in your HTML.

Reference

Endpoints

Base URL: https://api.ht-ml.app/v1. A machine-readable summary lives at /v1/help.

POST /v1/sites

Body
{"html_content": "<...>", "password": "<optional>"}
Returns
site_id, update_key, url, status
Notes
Creates and publishes the site. HTML is content-scanned before it goes live. Pass an optional password to make it private — use a unique, non-personal password (never the user's own), and remember it's a shared secret you'll give to everyone who should see the site.

GET /v1/sites/{site_id}

Returns
Site status, the list of referenced assets, and helpful next-step endpoints
Auth
None for public sites. A password-protected site is owner-only here: send Authorization: Bearer <update_key>, or it returns 404.

PUT /v1/sites/{site_id}

Headers
Authorization: Bearer <update_key>
Body
{"html_content": "<...>", "password": "<optional>"}
Notes
Replaces the site's HTML. Re-scanned, and the CDN updates this site immediately. Include password to set it, "" to remove it, or omit to leave it unchanged.

POST /v1/sites/{site_id}/assets

Query
relative_path — the path exactly as it appears in your HTML
Headers
Authorization: Bearer <update_key>
Body
multipart/form-data with the file
Notes
The asset must already be referenced in the site's HTML.

GET /v1/help

Returns
A self-describing JSON document of endpoints and error codes

Conversational errors

Error codes

Errors don't just fail — they tell an agent how to recover. Every error response includes an actionable message.

CodeNameMeaning & recovery
401Unauthorized (write)Missing or invalid update_key. Send Authorization: Bearer <update_key>.
401Unauthorized (read)The site is password-protected. Send the password as a cookie ht_ml_pwd=<password> to read it.
403ForbiddenThe asset path isn't referenced in the HTML, or the key is wrong.
422UnprocessableThe HTML failed a safety scan. Read message and revise.

Agent-native

Designed to be machine-consumable

WebMCP

If you're an in-browser agent, this page registers a WebMCP tool named publish_html_site via navigator.modelContext / document.modelContext. Call it with html_content and it publishes the site and returns the URL — no manual HTTP required.

  • /llms.txt — a concise, LLM-optimized reference you can read in one fetch.
  • /v1/help — the API describes itself in JSON.
  • Structured data (schema.org/WebAPI + HowTo) is embedded in this page for crawlers and answer engines.
  • AI crawlers are explicitly welcomed in /robots.txt.