Skip to main content
Glama
README.md
<p align="center">
  <img src="assets/mira-icon.png" width="72" alt="Mira">
</p>

<h1 align="center">Video Transcript MCP Server for TikTok, Instagram Reels &amp; YouTube — Mira</h1>

<p align="center">
  A remote <b>MCP server</b> that lets <b>Claude</b>, <b>ChatGPT</b>, <b>Cursor</b>, <b>VS Code</b> and any MCP client watch video links from<br>
  <b>TikTok, Instagram, YouTube, Pinterest, X (Twitter) and Facebook</b>: voiceover transcript, on-screen text, caption,<br>
  a structured extract, and the original video download.
</p>

<p align="center">
  <a href="https://mira-ai-skill.pro"><b>Website</b></a> ·
  <a href="#connect">Connect</a> ·
  <a href="#tools">Tools</a> ·
  <a href="#faq">FAQ</a> ·
  <a href="https://github.com/Mira-AI-Tools/mira-skill">Claude skill &amp; plugin</a>
</p>

<p align="center">
  <img src="assets/mira-skill.jpg" alt="Claude watching an Instagram Reel through the Mira MCP server and returning the recipe from it" width="820">
</p>

## Why

AI assistants can't open TikTok, Instagram or YouTube links: they hit a login wall or see only the page title. Mira watches the video for them. It transcribes the voiceover, reads on-screen text from key frames, takes the caption and returns one full extract, so the assistant answers from the whole video instead of guessing from the title.

<table>
  <tr>
    <td align="center"><b>Without Mira</b></td>
    <td align="center"><b>With Mira</b></td>
  </tr>
  <tr>
    <td><img src="assets/chat-before.webp" alt="An assistant can't open an Instagram Reel link" width="360"></td>
    <td><img src="assets/chat-with-mira.webp" alt="With the Mira MCP server, the assistant returns the recipe from the same Reel" width="360"></td>
  </tr>
</table>

## What you get

- **Voiceover transcript**: what was said in the video, in any language
- **On-screen text**: captions, subtitles and text overlays read from the frames
- **Caption**: the description under the post
- **Structured extract**: places with addresses and ratings, recipes with ingredients and steps, products, books, movies, workouts, study notes
- **Original video file**: the MP4, or every photo and video of a carousel *(Extract + Video plan)*

## Supported platforms

| Platform | Links |
|---|---|
| **TikTok** | videos, photo carousels, `vm.tiktok.com` short links |
| **Instagram** | Reels, posts, carousels |
| **YouTube** | Shorts and regular videos, long videos too |
| **Pinterest** | video pins |
| **X (Twitter)** | posts with video |
| **Facebook** | public videos and reels |
| **Web pages** | articles, recipes, any public page |

## Server

| | |
|---|---|
| Transport | Streamable HTTP (remote, nothing to install or run) |
| URL | `https://mira-ai-skill.pro/mcp` |
| Auth | `Authorization: Bearer mira_sk_…` header, **or** your personal connector URL `https://mira-ai-skill.pro/mcp/<key>` with no header |
| Registry | `io.github.Mira-AI-Tools/mira` in the [MCP Registry](https://registry.modelcontextprotocol.io) |

Your key and personal connector URL arrive on the setup page and by email after you [subscribe](https://mira-ai-skill.pro). Keep them private: anyone with the URL can use your daily videos.

<h2 id="connect">Connect</h2>

### Claude (claude.ai, Desktop, mobile)

Settings → **Connectors** → *Add custom connector* → name it `Mira` and paste your personal connector URL. Start a new chat and paste a video link.

For better answers, also add the [Mira skill](https://github.com/Mira-AI-Tools/mira-skill): it tells Claude when to call the tools and how to use the extract.

### ChatGPT

Settings → **Apps & Connectors** → *Advanced settings* → enable **Developer mode** → *Create* → paste your personal connector URL, authentication: none. Custom connectors are available on ChatGPT plans that support developer mode.

### Claude Code

```bash
claude mcp add --transport http mira https://mira-ai-skill.pro/mcp \
  --header "Authorization: Bearer mira_sk_your_key"
```

Or install the plugin, which bundles the server and the skill and asks for the key in its settings:

```bash
/plugin marketplace add Mira-AI-Tools/mira-skill
/plugin install mira@mira-ai-tools
```

### Cursor

`~/.cursor/mcp.json` (or `.cursor/mcp.json` in a project):

```json
{
  "mcpServers": {
    "mira": {
      "url": "https://mira-ai-skill.pro/mcp",
      "headers": { "Authorization": "Bearer mira_sk_your_key" }
    }
  }
}
```

### VS Code (GitHub Copilot agent mode)

`.vscode/mcp.json`; VS Code asks for the key once and stores it securely:

```json
{
  "inputs": [
    { "type": "promptString", "id": "mira-key", "description": "Mira API key (mira_sk_…)", "password": true }
  ],
  "servers": {
    "mira": {
      "type": "http",
      "url": "https://mira-ai-skill.pro/mcp",
      "headers": { "Authorization": "Bearer ${input:mira-key}" }
    }
  }
}
```

### Windsurf

`~/.codeium/windsurf/mcp_config.json`:

```json
{
  "mcpServers": {
    "mira": {
      "serverUrl": "https://mira-ai-skill.pro/mcp",
      "headers": { "Authorization": "Bearer mira_sk_your_key" }
    }
  }
}
```

### Gemini CLI

`~/.gemini/settings.json`:

```json
{
  "mcpServers": {
    "mira": {
      "httpUrl": "https://mira-ai-skill.pro/mcp",
      "headers": { "Authorization": "Bearer mira_sk_your_key" }
    }
  }
}
```

### Other MCP clients (Cline, Codex, Zed, LibreChat, n8n, …)

Add a remote MCP server of type Streamable HTTP and use your **personal connector URL** as the server URL. It needs no headers, so it works in any client that accepts a URL.

<h2 id="tools">Tools</h2>

All tools are read-only: they never change anything on your machine or in your accounts.

| Tool | What it does |
|---|---|
| `parse_video` | Watches a TikTok, Instagram, YouTube, Pinterest, Facebook or X link (or a web page) and returns the transcript, on-screen text, caption and structured extract. On the Extract + Video plan, also a download link for the original video or every file of a carousel. Long videos return a `job_id` instead. |
| `get_result` | Returns the extract of a video that was still processing, by its `job_id`. |
| `get_account` | Shows your plan, today's usage and a link to manage or cancel the subscription. |

Short videos are ready in seconds; long ones take a few minutes and the assistant picks the result up with `get_result`.

## What you can ask

> *"What's the recipe in this Reel? Turn it into a shopping list."*

> *"List every place from this TikTok with addresses, I'm going to Tokyo next week."*

> *"Transcribe this YouTube Short, translate it to English and summarize it in 3 bullets."*

> *"Which products does this creator recommend, and what do they cost?"*

> *"Break down this competitor's ad: hook, structure, call to action."*

> *"Make study notes from this lesson."*

<h2 id="faq">FAQ</h2>

**Is the server open source?**
No. Mira is a hosted service: this repository holds the documentation and the registry entry (`server.json`), not the server code. There is nothing to install or run.

**Do I need an API key?**
Yes, from a Mira subscription. Without a key the tools still connect and only explain how to subscribe; no video is processed.

**How many videos can I watch?**
20 videos a day on each plan. The same link asked again on the same day doesn't count twice.

**Which videos work?**
Most public videos. Private, deleted or region-locked videos may not be available, because access depends on each platform.

**What happens to the videos I send?**
Video files, transcripts and extracts are deleted from Mira's systems within 21 days. See the [Privacy Policy](https://mira-ai-skill.pro/privacy.html).

**I lost my key or connector URL.**
Get a new one at [mira-ai-skill.pro/recover](https://mira-ai-skill.pro/recover) with the email you subscribed with.

---

Mira works with a Mira subscription — see [mira-ai-skill.pro](https://mira-ai-skill.pro).
[Terms](https://mira-ai-skill.pro/terms.html) · [Privacy](https://mira-ai-skill.pro/privacy.html) · [Refunds & cancellation](https://mira-ai-skill.pro/refund.html) · Support: support@mira-ai-skill.pro

Documentation in this repository is released under the [MIT License](LICENSE).