Use these docs with an AI

This documentation is built to be read by machines as well as people. Every page has a plain-Markdown twin, the whole corpus is available as text bundles, and there are two MCP servers — one for the docs, one for the API itself.

Copy any page

Every page header carries a Copy Markdown button and an Open ▾ menu:

Markdown twins

Append .md to any documentation URL:

curl https://docs.theonlyapi.com/docs/webhooks.md
curl https://docs.theonlyapi.com/docs/crm/messaging/send-a-dm-or-ppv.md

Content negotiation works too — request the normal URL with Accept: text/markdown and you get Markdown back:

curl -H 'Accept: text/markdown' https://docs.theonlyapi.com/docs/pagination

Text bundles

The split exists for a practical reason: llms-full.txt is over a megabyte, which is more than most models want in one paste. llms-guides.txt is the file a human should actually hand to an assistant.

# Give an assistant everything it needs to write an integration
curl https://docs.theonlyapi.com/llms-guides.txt

MCP servers

There are two, and they do different jobs. Connecting the wrong one is the usual confusion.

Docs MCP

Read-only access to this documentation. No account, no key.

claude mcp add --transport http theonlyapi-docs https://docs.theonlyapi.com/mcp

{
  "mcpServers": {
    "theonlyapi-docs": {
      "type": "http",
      "url": "https://docs.theonlyapi.com/mcp"
    }
  }
}

Tools:

Use this when you want an agent to learn the API — answer questions, write integration code, look up a parameter.

API MCP

The product's own MCP server, which actually calls the API on your behalf: 59 tools across accounts, fans, messaging, earnings, subscribers, campaigns, webhooks, automations and events.

claude mcp add --transport http theonlyapi https://theonlyapi.com/mcp

It supports OAuth 2.1 with PKCE, and clients that cannot do OAuth may send a CRM API key as Authorization: Bearer <key>.

Notable behaviours:

• The destructive and money-moving tools require confirm=true — sending a DM, requesting a payout, changing subscription price, deleting an account, webhook or automation, and any non-GET through the escape hatches. Lower-risk writes (tagging a fan, creating a webhook or automation) apply immediately.
• Non-GET proxying is off by default. Enable mcp_unsafe_proxy in the dashboard danger zone, or via PATCH /api/crm/{crm_id}/mcp/unsafe-proxy, if you need the escape-hatch tools. Note the toggle applies to static bearer-key clients only — OAuth sessions (ChatGPT, Claude.ai) never get non-GET proxying.
• No dedicated vault, media-upload, mass-DM or scheduling tools exist — of_send_message sends one text DM to one fan, with no price, media or recipient list. Those surfaces are reachable only through the of_crm_request escape hatch, which needs confirm=true for any non-GET.

Full setup instructions for ChatGPT, Claude, Cursor and Claude Code are on the MCP integration page.

The OpenAPI spec

If your tooling speaks OpenAPI, skip the prose entirely:

curl https://theonlyapi.com/api/openapi.json

OpenAPI 3.1, 450 paths, 529 operations, every operation with a summary. This reference is generated from it.

Prompting suggestions

• Give it the guides, not the full bundle. llms-guides.txt plus the two or three endpoint pages you care about beats a megabyte of everything.
• Name the surface. "Use the CRM layer, not the /api2/v2 passthrough" removes most wrong answers, because the two have different response envelopes.
• Warn it about the money rule. Agents will happily write SUM(amount) for earnings, which is wrong — see the chargeback trap.
• Warn it about pagination. The /chats and /messages quirks in Pagination are not guessable; an agent will write a walker that loses data unless it has read that page.