PiratePressPiratePress

PiratePress MCP Server: Video Generation from AI Agents

The PiratePress MCP server is a way to hand video production to an AI agent: you write "make a video about …", and the agent orders it, waits and collects the mp4 itself. Here is what MCP is, which tools the server provides, and how to install it in a minute.

What you need to prepare

  1. An AI agent with MCP support — Claude Code, Kimi Code, Cursor, Claude Desktop, or any other Model Context Protocol client.
  2. An API key. The bot @piratepress_bot → /apikey. The pp_… key is shown once.
  3. One minute for installation — the one-liner does everything.

How to connect and use it

Step 1. Install with one command:

curl -fsSL https://piratepress.fun/install.sh | PIRATEPRESS_API_KEY=pp_... bash

The installer places the server in ~/.piratepress/mcp/, registers it in Claude Code (if installed), and installs the skill via the skills CLI with auto-detection of your agents. Nothing outside $HOME is touched. Manual installation — the @piratepress/mcp npm package, with client configs in the README.

Step 2. Meet the tools. The main ones:

  • generate_video — the recommended default: explicit parameters (the full POST /videos surface: theme, lang, duration, bg_ai, music, caption_mode, placement, cta, count, and more). Returns {id, cost, eta_seconds}.
  • quick_video — one free-form prompt; an LLM maps it to parameters. Only for vague one-off orders.
  • get_video_status — order status; on done it returns result_url and the posting pack.
  • wait_video — blocking wait for the finale (polls every 20 s with progress notifications, 30-minute default timeout).
  • review_video — script review decision in director mode: approve / edit / regen.
  • upload_asset / list_assets — the media library: custom backgrounds, tracks, voice samples (the returned id goes into *_asset_id fields).
  • list_bg_presets — valid stock background ids for bg_preset.
  • get_balance — wallet: dublons, subscription, fair-use remainder.

Step 3. Hand the agent a task. "Make five 30-second videos about space in English" — the agent checks the balance, creates orders via generate_video, waits via wait_video, and downloads the mp4s with posting packs.

Common mistakes

quick_video as the default. A free-form prompt is convenient, but the LLM mapping can drift off-topic, the price is known only after creation, and the mapper is throttled to a few calls per hour. For pipelines — generate_video with explicit parameters.

Ordering without a balance check. An agent that immediately queues 20 videos hits insufficient_funds. The right flow: get_balance → cost estimate → order.

Ignoring refunded. On generation failure dublons are returned and the order goes to refunded. The agent must distinguish it from error and know how to re-order.

FAQ

Why MCP when there's a REST API? REST is for your code; MCP is for ready-made AI agents: zero integration — tools appear in the agent right after installation.

Do I need a separate server? No — the MCP server is a local process on your machine, calling the public API with your key.

What about key security? The key is stored locally in the agent's config. Re-issuing it in the bot revokes the old one automatically.

Is director mode supported? Yes: director_mode pauses the order at a script review, and the agent decides via review_video.

Can a team use the server? Yes: everyone installs locally with their own key. Balances are per account — the key is tied to the account, not the machine.

What if the agent doesn't see the tools? Restart the client after installation: MCP registration is picked up at startup. You can verify with your client's server-list command.

Are there limits on parallel orders? The generation queue is shared per account: batches up to 50 videos per order are the normal mode. For extreme volumes, spreading orders over time is sensible.

Connection takes a minute: grab the key in @piratepress_bot with /apikey and run the one-liner — from there the agent makes the videos.