Give your AI agent your brand voice, not a guess at it.

Connect it

One endpoint and a token. Add this to your MCP client's configuration:

{
  "mcpServers": {
    "mora": {
      "type": "http",
      "url": "https://app.mora-marketer.com/api/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_MORA_TOKEN"
      }
    }
  }
}

Works with Claude Desktop, Claude Code, Cursor, and any client that speaks Streamable HTTP.

The five tools

Brand voice, as attachable context

Tools are fetched when a model decides it needs something. A resource is attached by you, once, and stays in context for the whole conversation. For brand voice that difference is the point: an agent that might look up how your business sounds will sometimes not bother, and what it writes when it does not bother is exactly the confident, fluent, off-brand copy you were trying to avoid.

@mora:mora://brand/voice   # Claude Code
Add Context > MCP Resources   # VS Code

It renders as prose, not JSON: how you sound, words to use and avoid, what never to write about, what you actually sell and for how much. If your brand kit is empty it says so and tells the agent not to invent one.

Prompts that carry the rules

Three named workflows, invoked as slash commands. They arrive carrying constraints an agent would not otherwise know — asking for upvotes is against Product Hunt's rules and can penalise your listing, and “Coming Soon” pages were retired in August 2025 while most launch guides still on the web recommend building one.

/mcp__mora__product_hunt_launch_week   "Tuesday 12 August"
/mcp__mora__first_maker_comment
/mcp__mora__write_like_what_worked

Prompts and resources are supported in Claude Code and VS Code today, and in Claude Desktop. ChatGPT's connectors are tools-only, which is why every capability here is also reachable as a tool — nothing is locked behind a primitive your client may not speak.

What it deliberately cannot do

This is a design decision, not a roadmap gap. An MCP client is an autonomous agent, and an agent operating on a real business's social presence without a person watching is not something we are willing to ship yet.

• —Publish anything, anywhere.
• —Schedule a post, or change one that is scheduled.
• —Connect, disconnect or modify a social account.
• —Change your billing or plan.
• —Write to your brand kit or brand voice.
• —Read the Discover article feed — that content is licensed from third parties and is not redistributed here.

Authentication

Send your Mora session token as Authorization: Bearer <token>. It is checked on every message, not once per session. A 401 carries a WWW-Authenticate header pointing at our Protected Resource Metadata (RFC 9728).

Being straight about a gap: Mora does not yet run an OAuth 2.1 authorization server, so the metadata document lists no authorization servers rather than pointing you at one that does not exist. Bring a token from the app. OAuth is the next thing we build here.

Limits and behaviour

• Every tool takes a limit, capped at 50, defaulting to 10–20.
• Results are scoped to your account. A team member sees their account's data, not their own empty slice.
• A very large result is truncated with a message telling the agent to narrow the query — never silently cut.
• The server is stateless: no session ids, no event stream. GET and DELETE return 405.

Troubleshooting

401 Unauthorized

Your token is missing or expired. Session tokens are short-lived — fetch a fresh one.

403 Origin not allowed

A browser-based client sent an Origin we do not allow. Native clients send none and are unaffected.

405 on GET

Expected. The server is POST-only; there is no event stream to open.

A tool returns hasBrand: false

No brand kit is set up on that account yet. Set one up in the app — the agent is telling you the truth rather than inventing a brand.