---
updatedAt: 2026-08-20T15:17:32.000Z
agentTools:
  siteIndex: https://developer.bitmovin.com/llms.txt
  projectIndex: https://developer.bitmovin.com/platform/llms.txt
---

# AI at Bitmovin

Overview of Bitmovin's AI-powered products and how to connect the Bitmovin MCP server.

# AI at Bitmovin

Bitmovin is built for AI-augmented developers. Point your AI agent — Claude, Codex, Cursor, Windsurf, GitHub Copilot, ChatGPT — at Bitmovin and have it build, encode, and operate your video workflows directly. This page is the entry point.

| Surface                                                                 | What it gives your agent                                                                          |
| ----------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------- |
| **[Bitmovin skill](https://bitmovin.com/skill)**                        | One-prompt context on every Bitmovin product, installed into the agent's local context            |
| **[`mcp.bitmovin.com`](#the-bitmovin-mcp-server)**                      | Live, OAuth-scoped access to your encodings, playback sessions, analytics, stream tests, and docs |
| **[Bitmovin CLI](#bitmovin-cli--manage-everything-from-your-terminal)** | The same surface from your terminal — scripts, pipelines, CI                                      |

Beyond the agent experience, Bitmovin also ships AI products of its own: [AI Scene Analysis](#ai-scene-analysis) for video metadata, and the [Bitmovin Assistant](#bitmovin-assistant) for in-dashboard chat.

## Bitmovin skill — drop us into your AI agent

[bitmovin.com/skill](https://bitmovin.com/skill) is the one-prompt way to give any AI coding agent context on Bitmovin's full product suite. Two paths:

**Paste into your AI:**

> Learn about Bitmovin from bitmovin.com/skill

**Or run in your terminal:**

```shell
npx @bitmovin/skills
```

The wizard detects your AI tool (Claude Code, Codex, Cursor, Windsurf, Copilot) and installs the canonical Bitmovin skill into the right place. After that, your agent knows how to use Player, VOD/Live Encoding, Observability, AI Scene Analysis, Streams, and Stream Lab — and will walk you through connecting the MCP server below.

Source: [github.com/bitmovin/skills](https://github.com/bitmovin/skills).

## The Bitmovin MCP server

We expose Bitmovin over the [Model Context Protocol (MCP)](https://modelcontextprotocol.io/), so you can connect it to Claude, ChatGPT, Cursor, or any other MCP-compatible client and use Bitmovin alongside your other tooling.

There is **one endpoint**:

```
https://mcp.bitmovin.com
```

One URL, one sign-in, every Bitmovin product. Tools are namespaced by product so your agent can pick the right one without name collisions.

### What's inside

| Product area  | Tool prefix       | What your agent can do                                                                      | Details                                                                                                    |
| ------------- | ----------------- | ------------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------- |
| Documentation | `general_docs_*`  | Answer questions from the developer docs, fetch a specific page, search public SDK examples | [Documentation tools](./mcp-documentation)                                                                 |
| Encoding      | `encoding_*`      | Inspect and debug your VOD and live encodings, explain failures, summarize activity         | [Encoding tools](./mcp-encoding)                                                                           |
| Player        | `player_*`        | Render a real Bitmovin Player inline in the conversation                                    | [Player tools](./mcp-player)                                                                               |
| Observability | `observability_*` | Query playback analytics — metrics, filters, impressions, licenses                          | [Observability MCP Server](https://developer.bitmovin.com/playback/docs/bitmovin-observability-mcp-server) |
| Stream Lab    | `streamlab_*`     | Create streams and test environments, run automated playback tests, read results            | [Stream Lab MCP Server](https://developer.bitmovin.com/playback/docs/bitmovin-stream-lab-mcp-server)       |
| Support       | `support_*`       | Create and look up Bitmovin support tickets                                                 | —                                                                                                          |

Which tools you actually see depends on what your API key or account has access to.

### Authentication

**OAuth is the default and the recommended path.** Add the server URL and your client walks you through a browser sign-in; no key handling, and access is scoped to your Bitmovin account.

For headless use — CI, scripts, shared automation — you can present a [Bitmovin API key](https://dashboard.bitmovin.com/account/api-keys) instead.

> **Send an API key as `Authorization: Bearer <key>`, not as `x-api-key`.**
> The hosted endpoint sits behind an edge that requires an `Authorization` header on every request and will reject an `x-api-key`-only request with `401` before it reaches the server.

**Multi-org accounts:** add `x-tenant-org-id: <id>` alongside your key to target a sub-organization. Note that the organization is fixed for the lifetime of a connection — to switch orgs, reconnect (and, on the OAuth path, sign in again).

## Connecting

Three ways in, depending on how you work.

### 1. Command line (Claude Code and similar)

```shell
claude mcp add bitmovin --transport http https://mcp.bitmovin.com
```

Run `/mcp` inside Claude Code to complete the browser sign-in, and `claude mcp list` to verify.

To use an API key instead of OAuth:

```shell
claude mcp add bitmovin --transport http https://mcp.bitmovin.com \
  --header "Authorization: Bearer $BITMOVIN_API_KEY"
```

### 2. Desktop config file

Claude Desktop's config file starts MCP servers as local processes, so point it at the server through [`mcp-remote`](https://www.npmjs.com/package/mcp-remote), which bridges a remote HTTP endpoint to a local one.

Edit `~/Library/Application Support/Claude/claude_desktop_config.json` (macOS) or `%APPDATA%\Claude\claude_desktop_config.json` (Windows) and add:

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

Then fully quit and relaunch Claude Desktop. On first use, `mcp-remote` opens a browser to sign you into Bitmovin and caches the tokens.

To use an API key instead:

```json
{
  "mcpServers": {
    "bitmovin": {
      "command": "npx",
      "args": [
        "-y",
        "mcp-remote",
        "https://mcp.bitmovin.com",
        "--header",
        "Authorization:${BITMOVIN_AUTH}"
      ],
      "env": {
        "BITMOVIN_AUTH": "Bearer YOUR_BITMOVIN_API_KEY"
      }
    }
  }
}
```

> Note there is **no space** after `Authorization:`. Claude Desktop on Windows and Cursor don't escape spaces inside `args` when launching the process, which mangles the header value — passing the value through `env`, where spaces are safe, avoids it.

The same shape works for Cursor, Windsurf, and other clients that launch MCP servers as local processes. Check your client's MCP documentation for the exact file location.

### 3. Custom connector (Claude organization settings)

Best for teams: an admin adds Bitmovin once and everyone in the organization gets it, with no local install and no API keys to distribute.

1. In Claude, go to **Settings → Connectors**. Organization owners and admins will also see an organization-level connector section.
2. Click **Add custom connector**.
3. Enter the URL `https://mcp.bitmovin.com` and save.
4. Each member clicks **Connect** and signs in with their Bitmovin account. Access is scoped to that member's own Bitmovin permissions.

Because this path is OAuth-only, it doesn't carry custom headers — multi-org users targeting a sub-organization should use one of the header-based options above.

### ChatGPT

Bitmovin is available as an official app in the [ChatGPT app directory](https://chatgpt.com/apps) — no MCP URL to copy:

1. Go to [chatgpt.com/apps](https://chatgpt.com/apps).
2. Search for **Bitmovin**.
3. Click **Connect**.
4. Sign in with your Bitmovin account — you can start chatting immediately.

## Bitmovin CLI — manage everything from your terminal

The [Bitmovin CLI](https://github.com/bitmovin/cli) is the terminal companion to the dashboard and the MCP server. It covers encoding templates and jobs, inputs/outputs, codec configurations, player licenses, and Analytics queries — ideal for scripting, automation, and CI/CD pipelines.

```shell
npm install -g @bitmovin/cli
bitmovin config set api-key YOUR_API_KEY
```

What it covers:

* **Encoding templates** — create, validate, start, and monitor YAML-based workflows
* **Encoding jobs** — list, start, stop, and monitor with live progress
* **Inputs/Outputs** — manage S3, GCS, Azure, and HTTP storage connections
* **Codec configs** — list and inspect H.264, H.265, AV1, VP9, and audio configurations
* **Player licenses** — list and manage license keys and allowed domains
* **Analytics** — query playback metrics, impressions, and license info

Requires Node.js 20+. Public Beta. Full command reference and install instructions in the [`bitmovin/cli` repo](https://github.com/bitmovin/cli).

## AI Scene Analysis

[AI Scene Analysis (AISA)](https://developer.bitmovin.com/encoding/docs/ai-scene-analysis) is Bitmovin's flagship AI product. It transforms video content into rich, structured metadata by analyzing every scene for context, themes, and visual characteristics — making every content minute discoverable, reusable, and advertisable.

AISA runs as part of your encoding workflow and produces per-scene metadata including:

* **Scene boundaries** with start/end timestamps
* **Summaries and descriptions** of each scene
* **Visual elements** — objects, brands, settings, locations, characters
* **Mood and atmosphere** — lighting, time of day, weather, sentiment
* **IAB taxonomy classifications** and keywords for contextual ad targeting
* **Sensitive topic flags** and content ratings
* **Multi-language support** — analyze in any language, output in multiple languages

### Use cases

* **Intelligent ad placement** — automatic SCTE marker insertion and keyframe placement at natural scene boundaries, with contextual ad matching via IAB taxonomies. Integrates with AWS MediaTailor, Broadpeak, and SpringServe.
* **Content discovery** — feed scene-level metadata into recommendation engines and search.
* **Operational automation** — highlights extraction, metadata generation, subtitle and caption workflows.
* **Player enhancements** — contextual overlays and interactive features powered by scene metadata.
* **Performance analytics** — correlate viewer engagement with scene context to identify what works.

To get started, see the full [AI Scene Analysis documentation](https://developer.bitmovin.com/encoding/docs/ai-scene-analysis).

## Bitmovin Assistant

The [Bitmovin Assistant](./bitmovin-assistant) is a chat-based AI assistant built into the [Bitmovin Dashboard](https://dashboard.bitmovin.com). It gives you a single conversation surface for navigating the product, inspecting your encodings, searching documentation, finding SDK examples, and querying your Observability data.

No setup required — sign in and start asking questions at [dashboard.bitmovin.com/assistant](https://dashboard.bitmovin.com/assistant).

## Under the hood

* The **Bitmovin skill** is plain markdown served from [bitmovin.com/skill](https://bitmovin.com/skill) and installed into your agent's local context — no model or service runs on Bitmovin's side for skill use.
* The **MCP server** does not ship any AI model. It is a tool server — your chosen MCP client's model does the reasoning. `mcp.bitmovin.com` validates your OAuth token or API key, namespaces tools by product, and forwards each call to the service that owns it.
* **AI Scene Analysis** runs during encoding and produces structured metadata — no separate service to manage.
* The **Bitmovin Assistant** uses its own LLM behind the scenes — see [Bitmovin Assistant](./bitmovin-assistant) for details.

## Data handling

* The **Bitmovin skill** is static markdown — no user data, no requests back to Bitmovin when an agent reads it.
* The **MCP server** doesn't persist user data. It verifies your credential, forwards the call, and returns the result.
* **Documentation tools** use only public Bitmovin documentation and public SDK example repositories, and need no account.
* **Encoding, Observability, Stream Lab, and Support tools** scope every call to the credential you present — they can only see data your account already has access to.
* **Player tools** are stateless — they accept a stream URL and player config from the chat client and return a rendered player; your stream URLs are not persisted.
* **AI Scene Analysis** processes your video content during encoding. The generated metadata is stored with your encoding output and is accessible via the Bitmovin API.