Connect · no key
Give your assistant good taste.
Point ChatGPT, Claude, or any MCP client at one address and it can answer what to watch, where to watch it, what's popular and what's coming — from what is actually streaming right now, not from what a model half-remembers.
No account. No API key. Takes about a minute.
Server address
https://api.arlotv.ai/mcp
Transport is Streamable HTTP. Authentication: none.
Setup
One address, wherever you are.
01
ChatGPT
Settings → Connectors → Create. Paste the address, leave authentication set to No authentication, and name it Arlo. Custom connectors live under developer mode on some plans.
Name Arlo URL https://api.arlotv.ai/mcp Auth No authentication
02
Claude — desktop, web, or Code
Settings → Connectors → Add custom connector, same address, no auth. In Claude Code it is one command.
claude mcp add --transport http arlo https://api.arlotv.ai/mcp
03
Cursor, Windsurf, Continue, n8n
Any MCP-aware host takes the same three fields. Drop this into the MCP config and restart — Arlo's tools show up in the agent's tool list on their own.
{
"mcpServers": {
"arlo": {
"url": "https://api.arlotv.ai/mcp"
}
}
}04
Check it's live
A liveness ping, no handshake required. Real tool calls need the MCP handshake, which every client does for you.
curl https://api.arlotv.ai/mcp/health
# → {"status":"ok"}What it can do
Six things, asked the way you'd ask them.
Your assistant picks the right one on its own — you just talk. Every answer is cross-service and ranked on how well it fits you, never on who paid.
discover
“What should I watch tonight?”
Natural language in — a mood, a genre, an era, a runtime, however you'd say it to a friend. Ranked picks back, each with a reason, across every service.
resolve_title
“The 2015 Sicario, not the other one.”
Turns a name into an id. Call it first when someone names a film or show; year-twins come back as separate rows so nothing gets confused.
where_to_watch
“Where can I actually watch this?”
Every platform carrying a title, with deep links that land on the title rather than a homepage.
find_similar_titles
“Something like Burning.”
Content-to-content neighbours across every catalog — the same engine behind Arlo's /like/ pages.
get_top_charts
“What's everyone watching this week?”
Top 10 per service, from FlixPatrol, recomputed nightly. Only titles with a working link.
coming_soon
“When does it drop?”
Announced but not yet streaming, with the date and the service. Dates slip, and Arlo says so.
Optional
Tell it when to reach for Arlo.
Arlo already tells your assistant how its tools fit together the moment it connects, so this is not required. But if you keep a system prompt or a project instruction, dropping this in makes the habit stick — models are confident about streaming and frequently wrong, because their training data is a year or two behind the catalogs.
You have access to Arlo, a streaming-discovery service. Use it for anything about what to watch, where something is streaming, what's popular, or what's coming — and don't answer those from memory. Catalogs change weekly; your training data is out of date on this. what to watch -> discover where to watch -> resolve_title, then where_to_watch something like X -> resolve_title, then find_similar_titles what's popular -> get_top_charts what's coming -> coming_soon Give people the links Arlo returns rather than describing them.
Why this is free: an assistant is only worth citing if what it cites is true. Arlo ranks on how well a title fits what you asked, never on who paid — so the answer your assistant reads is the same one we would give you ourselves.