OpenAI API and MCP

Both use an API token (Users → API tokens) sent as Authorization: Bearer wuppy_…. They run with the token’s permissions.

OpenAI-compatible API

The hub serves /api/v1/models and /api/v1/chat/completions in the OpenAI format, with streaming. The model is the agent: assistant, site-care, reporter or editor. This API is read-only: writing tools are not available.

import OpenAI from "openai";

const client = new OpenAI({ baseURL: "https://hub.example.com/api/v1", apiKey: "wuppy_..." });
const reply = await client.chat.completions.create({
  model: "site-care",
  messages: [{ role: "user", content: "How are my sites doing?" }],
});

System messages are passed to the agent as extra context.

MCP server

POST /api/mcp is a Model Context Protocol server. For example with Claude Code:

claude mcp add --transport http wuppy https://hub.example.com/api/mcp \
  --header "Authorization: Bearer wuppy_..."

Tool	Does
list_sites, check_site, site_overview	Sites you can reach, their connection and a summary
get_context	The agency playbook and client brief
list_collections, list_content, read_content	Read content
site_activity	Recent changes
site_report, security_scan	Monthly report data and the security scan
create_draft, update_content	Write content (drafts only for new entries)
email_site_member	Email a member of a site (recorded in the audit log)

MCP accepts only Bearer tokens, never browser cookies.