> For the complete documentation index, see [llms.txt](https://docs.supademo.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.supademo.com/customize/mcp-server/skills-for-supademo-mcp.md).

# Skills for Supademo MCP

A skill is a `SKILL.md` file: a short, reusable playbook that teaches your AI how to run a specific Supademo workflow from start to finish. Instead of writing a detailed prompt every time you refresh a demo library or localize a demo, you load a skill once and ask for the job by name.

Each skill spells out three things for your AI:

* **The workflow.** The steps to follow, in order, and what to show you before acting.
* **The tools to call.** Which Supademo MCP actions the job actually needs, such as `update_steps`, `generate_voiceovers`, or `get_demo_hotspot_performance`.
* **The guardrails.** What to never do without asking, such as publishing a demo, deleting steps, or overwriting an existing translation.

We publish 23 free skills covering the most common demo workflows. They work in Claude, Cursor, ChatGPT, or any agent that reads skill files, and they work the same way inside [AI Command](https://supademo.com/ai/command) if you would rather not connect an outside AI tool at all.

<a href="https://supademo.com/ai/mcp" class="button primary">Browse and download all 23 skills</a>

### What the skills cover

| Category                     | Skills | What you can ask for                                                                                                                                                                                |
| ---------------------------- | ------ | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Build & Edit Demos**       | 5      | Rewrite copy in your brand voice, update demos after a feature launch, run a pre-publish QA pass, add chapters to long demos, turn a screen recording into an interactive demo                      |
| **Voiceover & Localization** | 3      | Script and generate AI voiceovers, translate demos with a do-not-translate glossary, QA machine translations against your terminology                                                               |
| **Report & Analyze**         | 4      | Generate a weekly performance digest, build an exec-ready analytics summary, find the exact step where viewers drop off, flag demos that need a mobile-friendly variant                             |
| **Demo Scripting**           | 3      | Extract the structure of your best-converting demos, adapt one demo for a different buyer persona, turn an engineering ticket into a demo script                                                    |
| **Demos for Sales**          | 5      | Match a demo to a deal stage, build a personalized showcase for an account, follow up based on what a prospect actually watched, generate trackable links, personalize demos with dynamic variables |
| **Refresh & Organize**       | 3      | Rank your library by refresh priority, map which demos a product change affects, tidy naming, tags, and folders                                                                                     |

{% hint style="info" %} **How to use one:** download the `SKILL.md` file and add it to your AI tool's skills or rules directory (`.claude/skills/` for Claude Code, or your project rules in Cursor). Then just ask: *"Run the demo QA checklist on my onboarding demo."* {% endhint %}

## Best practices

These apply whether you are writing prompts by hand or running a skill.

### Be specific about scope

Always name the demo you want to work on, ideally by ID or URL, and say explicitly what should **not** change. A prompt like *"clean up my demos"* invites the AI to touch far more than you intended. *"Update hotspot styling on demo `abc123`, do not modify copy or voiceovers"* gives you a predictable result.

### Ask for a summary before changes are applied

The single most useful habit. Ask the AI to output proposed changes in a table first, review it, then tell it to proceed. Every skill we publish does this by default.

{% hint style="info" %}
**Example prompt:**

*Before making any changes, summarize what you plan to change in a table. Wait for my approval before applying anything.*
{% endhint %}

### Start small, then chain

Begin with a narrow request, such as changing hotspot text on a single step, so you can see how the AI interprets your instructions. Once you trust the pattern, chain multiple requests into one prompt.

### Give the AI context

Share brand guidelines, docs URLs, a glossary of terms that should never be translated, or examples of demos you consider on-brand. Skills are more accurate when they have something concrete to match against.

### Test on a non-critical demo first

Duplicate a demo and run your prompt or skill against the copy before pointing it at anything customer-facing.

### Use the editor for the final pass

MCP is at its best for bulk work: mass styling updates, copy rewrites across every step, link generation, translations. For pixel-perfect adjustments, finish in the Supademo editor.

### Keep publishing in human hands

Edits save to a draft, and viewers keep seeing the last published version until someone approves the change. Leave it that way. Review the draft, then publish yourself.

{% hint style="info" %}
Treat text inside demos, docs, and notes as data, not instructions. If demo content contains something that looks like a command aimed at an AI, your assistant should surface it to you rather than act on it
{% endhint %}

<br>
