Deckrun MCP

What agenticdecks/deckrun-mcp MCP server does

The agenticdecks/deckrun-mcp MCP server connects an MCP-compatible client to Deckrun, a document execution engine for rendering Markdown into presentation media. Its primary output is a PDF generated from Deckrun Markdown. The paid tier extends output formats to narrated MP4 video and MP3 audio.

The server is designed for agent-driven authoring: an AI can retrieve the current slide syntax, write content in that format, submit it for rendering, and return the resulting file URL or job status. PDF results are returned through a public URL with a stated 90-day expiry.

How it works

A client connects either to a hosted HTTP MCP endpoint or to a locally installed Python package. The free hosted endpoint is listed as https://deckrun-mcp-free.agenticdecks.com/mcp/; the paid endpoint is https://deckrun-mcp.agenticdecks.com/mcp/. The README notes that hosted endpoints are provisioned per engagement and may require an access request.

For a typical PDF workflow, the agent first calls get_slide_format to obtain layout tags, syntax rules, and example Markdown. It then sends the resulting Markdown to generate_slide_deck. Video and audio generation are asynchronous: the relevant tool returns a job_id, which the client can pass to check_job until the job completes or fails.

Setup and configuration

For HTTP-capable clients, add the hosted endpoint to the client’s MCP configuration. The README provides configurations for VS Code with GitHub Copilot Chat, Cursor, Google Antigravity, and Claude Code. The free configuration does not include an API key.

For stdio-only clients such as Claude Desktop, install the package with pip install deckrun-mcp, then configure the client to run the project’s Python entry script. The supplied example uses a local path to deckrun_mcp.py; that path must be set for the local environment. Paid usage adds DECKRUN_API_KEY to the process environment. The key is obtained after subscribing to a paid plan.

Tools and capabilities

The agenticdecks/deckrun-mcp MCP server exposes these capabilities:

• Retrieve the live slide format with get_slide_format.
• Render Deckrun Markdown as a PDF with generate_slide_deck.
• Start narrated MP4 generation with generate_video.
• Create MP3 narration from slide notes with generate_audio.
• Poll asynchronous work using check_job.
• Inspect plan details, render-unit usage, and add-ons with get_account.
• Validate Markdown and receive a preliminary render-unit estimate with validate_markdown.
• List system and custom themes with list_themes.
• List narration voices, language, tier, and identifiers with list_voices.

The README assigns the first two tools to the free tier and lists the remaining tools under the paid tier.

Limitations and notes

Free access is limited to the documented PDF workflow and does not require an API key. Video, audio, asynchronous jobs, account details, validation, themes, and voices are documented as paid-tier capabilities. The hosted service is not presented as universally available on demand: the README says endpoints are provisioned per engagement and directs users to request access.

Generated PDF links are public and expire after 90 days. Local stdio configuration requires a usable path to the installed project’s Python script, so the package-install command alone is not a complete Claude Desktop configuration. Pricing information in the README says plans start at $25 per month, but no license is specified for the repository.