Content

What the Content MCP server does

Content MCP server connects MCP-compatible agents and IDEs to the Content engine. The engine accepts URLs, uploaded or server-side files, and inline text, then determines which outputs can be produced from each source. Possible results include video, audio, subtitles, transcripts, summaries, translations, images, metadata, Markdown, and PDF.

Requests are declarative: the client describes the desired results instead of assembling separate download, media-conversion, transcription, and language-model steps. Content plans the work, runs the available processing tools, records the job history, and delivers the resulting artifacts. This makes the MCP interface suitable for agents that need to create or retrieve concrete files from source material.

The MCP server is one client of the same Content engine used by Content Studio, HomeTube, the CLI, the typed Python SDK, the Chromium extension, and the REST API. These clients use a shared contract and expose the same resolved capabilities, although their interfaces differ.

How it works

A source enters the engine as a URL, file, or text request. Content analyzes that source and resolves what it can produce before running the job. For example, a YouTube URL may produce a video, audio file, subtitles, transcript, summary, and translated or document-based outputs; a web page or Markdown file may produce text, a summary, a translation, and a PDF.

The Content MCP server gives an agent access to this workflow without requiring the agent to manage yt-dlp flags, FFmpeg pipelines, transcription glue, or language-model orchestration itself. The backend also keeps track of where artifacts came from and delivers finished files through the configured Content installation.

Setup and configuration

The documented installation path for the MCP client is uv tool install content-mcp. The engine itself is deployed with Docker Compose. The published deployment can be started by downloading the deployment Compose file and example environment file, then running docker compose up -d; building from source is also documented for development or for including the optional speech-to-text runner.

The default local stack includes the Content API and worker, Content Studio, HomeTube, and Content Console. The API is available at http://localhost:8010/docs in the default setup. Generated data, jobs, and artifacts remain under the installation directory by default, while CONTENT_DELIVERY_DIR_HOST can point delivery at another NAS or media-library location.

Tools and capabilities

The Content MCP server can expose requests involving:

• URL, file, and inline-text sources
• Media and document generation
• Transcript, subtitle, summary, and translation production
• Markdown and PDF output
• Artifact delivery and recorded job history

The exact outputs depend on what Content resolves as possible for the selected source and the capabilities available in the running engine.

Limitations and notes

Content is self-hosted, so an agent needs access to a running Content installation rather than only the MCP package. The README describes Docker Compose as the runtime prerequisite for the complete local stack. HomeTube is an optional client focused on YouTube; the MCP server is the general agent-facing entry point.

The provided material does not specify individual MCP tool names, authentication settings, or client-specific configuration. It identifies Claude, IDEs, and other MCP agents as target consumers, but does not document a fixed list of supported MCP applications. Output availability is source-dependent because the backend resolves capabilities before execution.