Claude MCP means using the Model Context Protocol, Anthropic's open standard, to connect Claude to external tools and data. You point Claude Code, Claude Desktop, or any MCP-aware agent at an MCP server, sign in once, and Claude can call that server's tools inside a normal conversation. No plugin, no glue script.

I build heydecks, an AI slide creator that agents call over REST or MCP, so I spend a lot of time wiring Claude to tools. MCP is the cleanest way I have found to give Claude a new capability. This post covers what Claude MCP actually is, then does the thing the title promises: it adds a make-me-a-deck tool to your Claude agent, with the exact commands, the tools the server exposes, and a real run from a prompt to a live deck and an editable .pptx.

#What is Claude MCP, and how does it work?

Claude MCP is Claude using the Model Context Protocol, the open standard Anthropic released in November 2024, to talk to systems outside its own context window. Anthropic pitched it as a single way to connect AI assistants to the places data already lives, so you stop writing a bespoke integration for every tool. The shorthand that stuck, and that shows up near the top of the results for this query, is a USB port for AI: one socket, many devices.

There are two halves to it. Claude is the host (the client); the thing it connects to is an MCP server. A server advertises a set of tools, each with a name, a one-line description, and an input schema. Claude reads that list, decides when a tool fits what you asked, calls it with arguments, and gets structured results back to reason over. The model never sees your credentials; the server runs the action and returns data.

Servers come in two shapes. A local server runs on your machine and talks over stdio, which is how the filesystem and many older community servers work. A remote server runs over HTTP and is reachable by URL, which is how hosted services connect. That difference matters in a minute, because Claude Code and Claude Desktop handle the two transports differently.

#What can you build with MCP servers in Claude?

Roughly anything with an API. People wire Claude to GitHub for pull requests, Postgres for live queries, Figma to write UI copy straight into design files, Blender to build 3D scenes, and Home Assistant to run the lights. Each of those is a separate server you add once and then call in plain language. If you want a survey of what is worth adding, I keep a running list of the best MCP servers for Claude.

Most of those servers fetch or change data. The category I care about here is different: servers that produce a finished artifact. A deck is a good test case, because the model is genuinely good at structuring an argument and writing the words, and bad at fighting a slide editor's layout engine. Hand the structure and copy to Claude, hand the rendering and the brand to a server, and you skip the part everyone hates.

#How do I add a make-me-a-deck tool to Claude with MCP?

The fastest path is one CLI command. heydecks runs a hosted MCP server at https://heydecks.com/mcp, and the CLI wires it into Claude Desktop, Claude Code, and Cursor in a single step:

npx heydecks install

That sets up OAuth, so there is nothing to paste. If you would rather add it by hand, the command depends on which host you use and which transport it speaks.

For Claude Code, connect over HTTP and authenticate in the browser:

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

Restart, and the first call opens a browser to sign in with heydecks. Under the hood that is standard OAuth 2.1 with PKCE, so no token ever touches your shell history. For a headless box or CI, where a browser sign-in is awkward, mint an mcp_ token in the dashboard and pass it as a header instead:

claude mcp add --transport http heydecks https://heydecks.com/mcp \
  --header "Authorization: Bearer mcp_your_token"

Claude Desktop is the awkward one. It speaks stdio only, so a remote server has to be bridged. mcp-remote does that and handles the browser sign-in for you. Open Settings, Developer, Edit Config, and add:

{
  "mcpServers": {
    "heydecks": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://heydecks.com/mcp"]
    }
  }
}

Restart the host either way, then confirm the server registered:

claude mcp list

You should see heydecks in the list. That is the whole setup.

#What tools does the heydecks MCP server give Claude?

Here is the part that surprises people: the server does not write your slides. It is a structured deck store, not a generator. Claude designs the deck and writes every word; the tools persist exactly what it sends, so the result sounds like you and not like generic filler. There is no second model writing copy behind the scenes.

These are the tools you will actually watch Claude call:

There are more, for versioning, reordering, and custom slide types, but those eight cover the build-a-deck path. The brand part matters: every export is locked to your colors, fonts, and logo by the Brand Kernel, which you can pull straight from a website with extract_brand_from_url.

#What does an end-to-end deck build look like over MCP?

You ask in plain language and watch the tool calls scroll by. Here is a prompt I use to demo it:

"List my slide templates, then create a deck called 'Q3 review' with a title, three stats, a process flow, and a CTA. Publish it when done."

Claude calls list_slide_templates first to learn the field schemas, then designs the structure and writes the copy itself. It calls create_deck to get a deck id, add_slides to build all of it in one shot, and publish_deck to make it live. A few seconds later you get a link like heydecks.com/d/q3-review, and from that page you can export the same deck to a PDF or a native, editable .pptx. The PowerPoint is real slides, not images on a canvas, so your team can open it and keep editing.

Point Claude at a URL and the same flow runs with one extra step: extract_brand_from_url pulls a company's look, Claude reads the page, and the deck comes back already on that brand. That is the run in the embed below.

heydecks is the AI slide creator that AI agents call over REST or MCP. From a prompt, markdown, or a URL it returns a live deck link, a PDF, and a native, editable PowerPoint, every export locked to your brand by the Brand Kernel.

#Is Claude MCP free, and what does it cost?

The protocol is free and open source; what you pay for is access to Claude and to whatever each server does on your behalf. Claude Code needs a paid Claude plan or API credits to run at all. The free Claude tier gives you roughly 40 short messages a day and no Claude Code, so MCP through Claude Code starts at the Pro plan.

The heydecks server has its own meter, and it is the same on every channel. A deck costs 100 credits whether you build it in the dashboard, over the REST API, or through MCP. Reading decks, editing them, reordering slides, and exporting to PDF or PPTX are all free. Two tools cost more because they call models for you: generate_image is 80 credits per image and extract_brand_from_url is 50.

#Skills, MCP, or the REST API: which should you use?

Skills and MCP solve different problems, and they work together rather than competing. A Skill teaches Claude a procedure or a body of knowledge, like a coding convention or a checklist; MCP gives Claude a live connection to an external system. A Skill can call MCP tools as part of its procedure. Despite the recurring "is MCP dead" posts, MCP is not going away. Some things people first shipped as servers have moved to Skills because Skills fit library knowledge better, but anything that needs a live connection to your data still wants a server.

For building decks specifically, heydecks gives you two doors. MCP is the one to use when you want Claude to design the deck slide by slide and write the words, which is most agent work. The REST API is the one to use when you want a deck from a single call: POST /v1/generate takes a prompt and returns a job, and heydecks does the generation. If your goal is to build a PowerPoint with Claude, MCP is the natural fit, since Claude is already the thing holding the conversation.

Want this in your own Claude? Add the server in one line:

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

Below is a deck heydecks built from a URL, rendered on a sample brand. Click through it.

See the MCP server to connect Claude, try the API for free if you would rather call REST, and the docs cover both end to end.

#Frequently asked questions

#Is Claude MCP free to use?

The Model Context Protocol is free and open source, so the standard itself costs nothing. You pay for the pieces around it: Claude Code needs a paid Claude plan or API credits, and any given server may charge for what it does. The heydecks server, for example, is 100 credits per deck and free to read, edit, and export.

#What is the difference between Claude Code and MCP?

Claude Code is the agent that runs in your terminal; MCP is how that agent gets new tools. Claude Code ships with built-in abilities like reading files and running commands, and MCP servers add the rest, from GitHub to a database to a deck builder. You add a server once with claude mcp add, and from then on Claude Code can call it.

#Does Claude Desktop support remote MCP servers?

Yes, but not directly over HTTP. Claude Desktop speaks stdio, so a remote server like heydecks connects through the mcp-remote bridge, which you set in claude_desktop_config.json. Claude Code and Cursor connect to the same HTTPS endpoint natively, no bridge needed.

#Can Claude make an editable PowerPoint over MCP?

Yes. With the heydecks server connected, Claude builds the deck and you export it from the deck page as a native .pptx, real editable slides rather than flattened images. Every export is locked to your brand by the Brand Kernel, and the same deck is also available as a live link and a PDF.