Beyond the browser
The idea: every app in this book runs in a browser tab. That is perfect for one file and one person. It is the wrong tool for 300 invoices every Monday, for an AI assistant that should "just fix this PDF", or for documents that must never leave your network. This chapter covers the other doors into the same engine.
What this chapter covers
| Page | Channel | In one line |
|---|---|---|
| Which channel when | all | A decision table: web, CLI, MCP, REST, self-host, desktop |
| CLI plugins | command line | One small downloadable tool per operation; chain them in a shell |
| MCP server | AI assistants | Claude, Cursor, VS Code call Documentize tools in conversation |
| REST API | HTTP | Submit a task, poll its status, download the result |
| Self-Host | your server | The same API on your own machine, for private data |
| Desktop | installed clients | Windows, Linux and terminal clients, optionally with a local engine |
| Plugins for developers | code | How the plugin contract works under the CLI tools |
| The how-to guides library | reading | Each operation, step by step, per channel |
How to read it
- Start with Which channel when. It takes two minutes and saves you from picking the wrong door.
- Jump to the channel you chose. Each page is self-contained: install, run, chain.
- Keep the how-to guides open as a reference. They show the same operation on web, CLI and MCP side by side.
Three things true in every channel
- Same engine, same apps. A compress is a compress whether you click, type or ask.
- Same allowance rules. Free users get a small daily number of operations; subscribers do not. A subscription you hold on the web carries over to MCP; the clients and the self-host use your key. Exact numbers change, so check the pricing page (verify).
- Asynchronous by nature. Heavy work is queued, runs on a worker, and you collect the result. The web page hides this with a progress bar; the other channels expose it as "task id, then status".