Creator data your agent can just call.
One URL turns the Creator Data API into tools your agent already knows how to use. No client library, no auth plumbing, no polling loop — paste the endpoint and ask for a creator.
https://api-app.lolly.com/api/public/v1/mcpConnected in one command.
Streamable HTTP, so every MCP client speaks it. Bring the API key you already have, or let the client run the OAuth flow and paste nothing.
First call to try: get_usage_and_limits. It orders nothing, charges nothing and is never rate limited — if it answers, you are connected.
Three tools. All read-only.
Each one is annotated read-only and non-destructive, so a client can call it without asking your user to confirm every request.
lookup_influencerLook Up Creator Profile
One creator by handle, on any of the five platforms. Starts the analysis, follows it and returns the profile.
platformusernamefields{
"platform": "instagram",
"username": "nike"
}lookup_influencers_bulkLook Up Many Creator Profiles
Up to 100 creators in one call, platforms mixed freely. One outcome per creator — a handle that resolves to "not found" never fails the call.
itemsfields{
"items": [
{
"platform": "instagram",
"username": "nike"
},
{
"platform": "tiktok",
"username": "khaby.lame"
}
]
}get_usage_and_limitsGet Remaining API Budget
How much request and analysis budget is left, and when it replenishes — so an agent can pace a large job instead of discovering its limits by being throttled.
None.
{}Your agent never writes a poll loop.
A full creator analysis takes about 70 seconds. The REST contract asks the caller to come back every few seconds until it lands; the tool absorbs that instead.
The call holds the request open
It starts the analysis if one is not already running, streams progress while it waits, and returns the profile if it lands inside the window.
status: in_progress, call_again: true
Not an error — a turn of a conversation. The analysis keeps running; the agent makes the identical call again.
The repeat rides the same analysis
It attaches to the work already running rather than ordering a second one, and returns the profile when it lands.
Repeating is free.
A repeat spends no request budget, no analysis budget and no credits, and it can never be rate limited — exactly like polling a REST lookup. Abandoning the call cancels the work, and cancelled calls are not charged either.
Sized for a context window.
A raw creator profile is roughly 8 KB of JSON. One of those is affordable; a hundred is not — and a bulk call is the whole point. So every profile a tool returns is compacted by default, identically on all five platforms.
Two overrides, one argument
fields: ["followers", "audience_countries"]Exactly the fields you name, and no other profile data.
fields: ["*"]The complete payload — byte for byte what the REST lookup returns for that creator.
Every result stamps which of the three it carries, so a caller always knows whether it is holding the compact shape, a selection or the full payload — and the compact shape is versioned, so a field leaving it is a visible change rather than a silent one.
{
"status": "ready",
"platform": "instagram",
"username": "janedoe",
"call_again": false,
"projection": "compact",
"projection_version": "1",
"profile": {
"influencer_name": "Jane",
"is_verified": true,
"categories": ["Creators & Celebrities", "Music & Songs"],
"followers": 286927835,
"engagement_rate": 0.34,
"average_likes": 972429,
"audience_countries": {"United States": "28%", "India": "14%"},
"audience_age_ranges": {"18-24": "46.7%", "25-34": "24.5%"},
"posting_activity": {"posts_per_week": 0.94}
}
}The same key, the same bill.
The MCP endpoint is not a second product with its own pricing. It authenticates, meters and logs through the same path the REST endpoints use.
A key you already have
A live lly_live_… key works immediately, with no authorization flow. Status, expiry, revocation and scopes are enforced exactly as on REST: the lookup tools need influencers:read and influencers:refresh together, and a key without both is not even offered them.
Or OAuth 2.1, with no key at all
Clients that authorize themselves — Claude.ai connectors, Claude Code — discover everything from the endpoint address: protected resource metadata, a first-party authorization server, mandatory S256 PKCE, and a consent screen that names the scopes and the host you return to. Tokens are bound to this endpoint and revocable from the consent screen.
Charged only on a definitive result
One credit for a Report analysis, 0.05 for a Quick one — charged when a profile or a confirmed no-result comes back. A progress turn, a cancelled call and a technical failure are never charged.
The budgets an agent can read
Currently 4 requests per second and 240 creator analyses per rolling 60 seconds, per credential. An agent never sees a response header, so get_usage_and_limits reports both — free, unthrottled, and it orders nothing.
Give your agent the data.
Paste one URL and the Creator Data API is a tool call. Talk to us about volume, or read the reference and start now.
Let's Talk
Need something the three tools do not cover?
Let's TalkFrequently asked questions
The Model Context Protocol is how an AI agent discovers and calls tools over HTTP. Point a client at our endpoint and it reads the tool catalogue itself — names, arguments, what each one costs — so nobody writes an integration first. The alternative is a REST client, an auth layer and a polling loop before the agent sees any data.
Any client that speaks Streamable HTTP MCP. We verify Claude Code, Claude.ai custom connectors, Cursor and VS Code. Three protocol revisions are served side by side — 2026-07-28 (stateless) plus 2025-11-25 and 2025-06-18 (session-based) — and your client negotiates; there is nothing to configure.
No. A live lly_live_… key authorizes MCP tool calls with no extra step, as long as it carries the influencers:read and influencers:refresh scopes — the same pair the REST lookup routes require. Clients that prefer to authorize themselves can use OAuth 2.1 instead and hold no key at all.
A full analysis takes about 70 seconds and always reaches a terminal outcome within 12 minutes, which is longer than any MCP client will hold a request. So the tool holds it for as long as it safely can, then returns status in_progress with call_again: true. Make the identical call again: it attaches to the analysis already running, costs nothing, and is never rate limited.
By default a documented compact projection — 29 top-level fields, the same shape on all five platforms, always present and null where a platform has nothing. Pass fields with the names you want for exactly those, or fields: ["*"] for the complete payload. There are 46 selectable fields in total; the omitted ones are things like growth series, full post objects and lookalikes.
The same. One credit per Report analysis, 0.05 per Quick one, charged only on a definitive result, from the same balance and against the same rate-limit budgets. Progress turns, cancelled calls and technical failures are free, and the MCP endpoint needs the same active Creator Data API subscription the REST endpoints need.
No. All three tools are reads, and each is annotated read-only and non-destructive so a client can present them as safe. There is no tool that writes, deletes or changes anything in your account.
One URL. Your agent does the rest.
Creator profiles, audience demographics and engagement for Instagram, TikTok, YouTube, Snapchat and X/Twitter — as tool calls.