Skip to main content
Replyful exposes a remote Model Context Protocol server. Connect it to an AI assistant and ask it to write, translate and publish help-center articles.
Transport: Streamable HTTP (protocol revision 2026-07-28, older 2025 clients are served too).

​
Signing in

Two credentials work:
  • OAuth (recommended, per user). Add the URL in your client and sign in with your Replyful account. Tokens are scoped to the organization you sign in to and require the org:manage permission. Claude and ChatGPT both discover the sign-in flow automatically.
  • API key (per organization). Send Authorization: Bearer rfl_live_… on every request. Create keys under Settings → Developers → API keys. Use this for CLI tools and Claude Code.

​
Claude.ai / Claude Desktop

Customize → Connectors → Add custom connector, paste the URL and choose Sign in now. Team and Enterprise admins add it under Organization settings → Connectors. To use an API key instead, choose No sign-in and add a request header authorization with the value Bearer rfl_live_… (request headers are in beta).

​
Claude Code

​
ChatGPT

Enable Developer mode in ChatGPT settings, add a connector with the URL and complete the sign-in. Writing tools require a plan that allows write actions.

​
Tools

Every write goes through the same rules as the admin editor: the language must be enabled on the help center, expectedVersion must match the current version, and nothing is public until published. Deleting, settings, languages and custom domains stay in the admin app. A typical prompt: “Write a Swedish version of every English article in our Docs help center and publish them.” The assistant lists the center, reads each English article, creates the Swedish translation with save_article_draft (expectedVersion null) and publishes it.

​
Rate limits

The MCP endpoint shares the API rate limit: 100 requests per minute per user or API key.