Palmier Pro starts at $0 for editing and jumps to $29/month or $69/month once you want AI generation. That pricing split matters more than the demo. The interesting part is not just that Claude can trim clips over MCP; it is that Palmier is separating the open editor from the metered AI layer, which is a much saner model than pretending everything is "AI-native" and somehow free.
The source walkthrough is good because it stays concrete. Palmier Pro is a macOS-only, Swift-native NLE that exposes a local MCP endpoint at http://127.0.0.1:19789/mcp. With Palmier open, Claude Code, Claude Desktop, Cursor, or Codex can connect and operate the same timeline you are looking at: read tracks, trim clips, reorder sections, split footage, and, if you are paying for credits, generate new assets directly onto the timeline.
That is the genuinely new bit. Most AI video products still force the old loop: prompt somewhere else, wait, download, re-import, then continue editing. Palmier collapses that into one project. If you build or advise on custom MCP servers or Microsoft Copilot and AI agents, this is worth paying attention to because it shows what agent tooling looks like when the host app is designed for it instead of retrofitted.
What is actually good here?
The best decision Palmier made is architectural: the timeline is the system of record, not the chat window. The MCP server is local, the editor is open source under GPLv3, and editing/export stay free. That lowers lock-in compared with the usual black-box SaaS editor.
The setup is also refreshingly simple:
claude mcp add --transport http palmier-pro http://127.0.0.1:19789/mcp
Or in Cursor:
{
"mcpServers": {
"palmier-pro": {
"type": "http",
"url": "http://127.0.0.1:19789/mcp"
}
}
}
That is the sort of friction level MCP needs if it is going to become normal outside demos.
Where I would be skeptical
There are three catches.
First, platform support is brutally narrow: macOS 26 on Apple Silicon only. For most engineering teams, that makes this a creator workstation experiment, not an org-wide standard.
Second, "open source" comes with an asterisk. The editor, MCP server, and in-app chat are open, but the generative pipeline is closed and paid. That is not automatically bad, but it does mean the value everyone will demo is the one part you cannot fully audit or self-host.
Third, the cost model is likely to surprise people. The visible subscription looks low enough at $29/month for Pro and $69/month for Max, but community feedback is already pointing at the real issue: credits drain fast. That is the classic AI tooling trap. The editor buyer and the generation bill payer are often not the same person, and nobody adds guardrails until after the first messy invoice. If you are doing this at team scale, treat it like any other AI and automation audit problem: who can trigger generation, how much, and with what approval path?
The security angle people will skip
An agent editing your local timeline is fine. An agent with generation capability and full project context is more sensitive than it sounds. Prompts, clip order, project metadata, and raw media context can all be commercially sensitive. I would not give any agent a blank check here.
In practice, I would want:
- separate paid accounts for experimentation versus production
- clear limits on who can invoke generation
- a review step before exports leave the workstation
- documented retention answers for prompts and generated assets
This is the same governance pattern I recommend in AI workflow automation: the fastest path is rarely the safest default.
My take
Palmier Pro looks genuinely promising, and more credible than most "AI video editor" launches because it solves a real workflow break: the handoff between generation and editing. But it is still early software, with narrow platform support and a business model that shifts the real value into a closed, metered layer.
So yes, I would test it. I would not roll it out casually, and I definitely would not let the phrase "free and open source" hide the part that will actually hit your budget.
