What counts as a credit?

One successful API call. If a request fails — expired session, upstream error — it isn't charged. Mass DMs are the exception: a mailing bills one credit per recipient, and you can preview the exact cost before sending.

How do I authenticate?

A Bearer token: Authorization: Bearer ofapis_sk_…. Create and revoke keys in the dashboard.

Can I manage multiple creators?

Yes. Link several accounts and address each by id under scoped routes. Account count scales with your plan.

Do you have an OpenAPI spec?

A complete OpenAPI 3.0 document and a live Swagger playground — point your generator at it for a typed client.

OnlyFans API for AI Agents (MCP)

ofapis is a plain REST API with a complete OpenAPI 3.0 spec — the exact format modern AI agents consume. You don't need a bespoke plugin: any MCP (Model Context Protocol) bridge that wraps an OpenAPI service, or any assistant that imports an OpenAPI schema, can drive OnlyFans actions through ofapis with a single Bearer token.

What MCP is, and how ofapis fits

MCP is an open standard for connecting AI clients (Claude Desktop, Claude Code, and a growing list of others) to tools over a local server. A client asks the server to list_tools, gets back a catalogue with typed inputs, and calls a tool when the model decides to. ofapis doesn't need a hand-written MCP server, because the OpenAPI spec already describes every operation, its parameters, and its responses. A generic OpenAPI→MCP bridge reads that spec and generates the tool catalogue for you.

Why it works out of the box

OpenAPI-native. Every public endpoint is in the spec at https://api.ofapis.com/v3/api-docs/public, so tools are generated, not hand-maintained. When ofapis adds an endpoint, the bridge picks it up.
Bearer auth. One Authorization: Bearer ofapis_sk_... header — no OAuth dance for the agent to manage. The credential stays in the bridge's environment, out of the prompt.
Metered + safe. One successful call is one credit; failed calls (expired session, upstream error) aren't charged, so an agent that retries won't run up a surprise bill.
Sessions stay alive. Accounts connect once and ofapis keeps the session warm, so there's no refresh logic for the agent to handle.

How a bridge turns the spec into tools

The bridge fetches the OpenAPI document.
Each path + method becomes one MCP tool, named from its operationId, with a JSON schema for its parameters.
When the model calls a tool, the bridge issues the HTTPS request to https://api.ofapis.com/api/public/v1, attaches the Bearer header, and returns the JSON response to the model.

Most bridges let you allowlist operations — expose read-only tools to an autonomous agent and keep sends behind a human step.

What agents build with it

Churn rescue — detect lapsing fans from the subscribers API and draft win-back DMs via the messaging API.
Whale watch — rank top spenders, notice when one goes quiet, and alert you.
Segment + send — build a fan segment, preview the cost, and trigger a mass DM.
Morning digest — summarise new notifications and chats each day.

Wire it up

Grab a token — start free.
Point your MCP bridge (or assistant) at the OpenAPI spec and set the Bearer header.
Follow the per-client guide: Claude · ChatGPT · Manus · or the AI assistants overview.

FAQ

Is there a first-party ofapis MCP server?

Not yet, and it isn't required. ofapis is REST + OpenAPI 3.0, which existing MCP bridges and GPT Actions consume directly. A first-party MCP package is on the roadmap to remove the bridge step, but you can wire up an agent today without it.

How does an OpenAPI→MCP bridge authenticate?

You give the bridge your ofapis_sk_... token in its environment or config, and it attaches Authorization: Bearer ... to every request. The token never enters the prompt or the model's context.

Will an agent that retries cost me extra credits?

No. Only successful calls are billed. A failed call — expired session or upstream error — costs nothing, so retry loops stay cost-safe. See pricing for the credit model.

Which AI clients can use this?

Any MCP-capable client (Claude Desktop, Claude Code) via a bridge, plus assistants that import OpenAPI directly such as ChatGPT's Custom GPT Actions. See the AI assistants overview to pick a route.

Can I restrict what the agent is allowed to do?

Yes. Bridges typically support allowlisting operations, so you can expose only read endpoints (chats, subscribers, notifications) to an autonomous agent and route write actions like mailings through an approval step.

Related: Claude · ChatGPT · Full API docs