Open-source video editor Wunjo Make can now get plugins written via Claude Code, Codex and Cursor

The author of the open-source video editor Wunjo Make has published a guide on Habr to building plugins. You can write them by hand or hand the job to one of three AI coding agents: Claude Code, Codex or Cursor. So that the agents don't improvise the project structure as they go, the repository now includes the builder's code and a set of skills for cloud language models. A finished plugin can call a local neural network or an external API, and the editor can be controlled through MCP.

What was added to the repository

The announcement names two artifacts: a builder and skills. Judging by the description, the builder handles plugin packaging, so developers don't have to dig into the editor's internals. Skills are pre-written instructions for the model. The agent loads them into its context and learns what a plugin consists of, how to build it and which conventions the project follows.

This is usually the stage where vibe coding breaks down. A model writes functions well but doesn't know someone else's architecture. If you give it a description of the plugin format and a working builder, the result can be verified by running the build, not just by reading the code.

Local models, API and MCP

The main feature of the setup is that a plugin's backend isn't tied to a single way of running. A plugin with a local neural network fits when video can't be sent outside, or when you don't want to pay for every request. An API-based plugin is simpler: it doesn't need a GPU, but it does need a key and an internet connection.

A separate line in the announcement covers control through MCP. Anthropic introduced the Model Context Protocol in November 2024, and it has since become a common way to connect external tools to models. For a video editor, this means an agent can do more than write a plugin: it can also call the editor's own functions. The announcement doesn't say which operations are available through MCP; for that, you need to check the article and the repository.

Key parameters

• Editor: Wunjo Make, open source.
• Ways to create plugins: by hand or through Claude Code, Codex, Cursor.
• What was added to the repository: builder code and skills for cloud LLMs.
• Plugin backend: local neural networks or an API.
• Control: through MCP.

Where to start

Start by cloning the repository and connecting the skills to the agent you already use. To test the setup, it's best to pick a small plugin with a single function, for example a wrapper around a model for processing audio or frames. That shows right away how closely the agent follows the project's rules and whether the build passes without manual fixes.

When choosing between a local model and an API, it's worth assessing three things in advance: the amount of VRAM, the sensitivity of the material and the cost of requests on long videos. Read any code the agent generates before installing it. The plugin runs on your machine and may get access to your project files.

One last point. This is a guide from the project's author, not an independent test. How mature the plugin API turns out to be, and how well agents handle complex plugins, will become clear once third-party developers start publishing their own work.