Claude.ai

claude.ai supports remote MCP servers as Custom Connectors. PresentFast hosts one at https://presentfast.com/api/mcp. No local install, no terminal — sign in once in the browser and the model can publish decks straight from a conversation.

Status: Stable
Transport: http
Auth: OAuth · DCR
Install: —

How do I open Custom Connectors in claude.ai?

In claude.ai, go to Settings → Connectors → Custom Connectors and click Add a custom connector. (This setting may be gated to certain plans — see the Anthropic Help Center if you don't see it.)

What URL do I paste into the connector?

Use the remote MCP endpoint and an obvious name:

Name: PresentFast
URL:  https://presentfast.com/api/mcp

Note

claude.ai discovers the OAuth authorization server automatically via PresentFast's RFC 9728 Protected Resource Metadata at /.well-known/oauth-protected-resource. You won't see any client-id field — that's intentional. claude.ai uses Dynamic Client Registration (RFC 7591) to provision itself.

How do I authorize claude.ai to access PresentFast?

claude.ai opens a popup walking you through PresentFast's standard OAuth consent screen:

Sign in to PresentFast (or accept the existing session if you're already logged in).
Review the requested scope (all — publish, list, rename, read analytics) and click Allow.
The popup closes and the connector flips to connected in claude.ai.

Every connector approval shows up in your Settings on PresentFast — you can revoke claude.ai's access there at any time.

How do I publish my first deck from claude.ai?

Start a new conversation and ask Claude to use the connector. A test prompt that exercises the publish path:

Try this prompt
Use the PresentFast connector. Write a short Markdown deck about three uses of large language models, three slides separated by ---, then publish it titled "LLM uses".

Claude will call publish_presentation with the inline content, the model receives the share URL in the response, and repeats it back to you.

Note

Tools the model has: publish_presentation, list_presentations, get_presentation_analytics, update_presentation. Same names as the local stdio server, but publish_presentation takes inline content + filename here (the remote server can't read files from your machine).

Slide-break conventions

PresentFast splits one Markdown file into multiple slides using any of three conventions. They can be mixed in the same document.

1. Horizontal rule

A line containing only --- marks a slide boundary. This is the default and matches most slide tools (e.g. Marp, reveal.md).

# Slide 1

Some content.

---

# Slide 2

More content.

2. Page-break comment

An HTML comment  on its own line is also a slide boundary. Useful when --- would conflict with frontmatter or inline horizontal rules. Always on, no query param needed.

# Cover

```text
visible ---
inside fenced code is ignored
```

<!-- pagebreak -->

# Next slide

3. H2 per slide (opt-in)

Pass ?theme=h2-slides on the upload URL (the AI tool does this for you when it detects an H2-heavy document) and every ## heading starts a new slide. Useful when you have an outline-style doc you didn't author for slides.

Note

All three conventions are code-fence aware: a ---, , or ## inside a fenced code block is left intact.

How do I pin a specific connector version?

The remote MCP server is versioned with the rest of presentfast.com — Anthropic's connector picks up changes the next time it tools/lists. There's no version-pin equivalent for the remote endpoint; if you need stable behavior, use the local @presentfast/mcp-server via Claude Code, Cursor, or Codex instead (those guides cover @version pinning).