Palmier Pro: An Open-Source Video Editor That Lets AI Agents Drive the Timeline

Palmier Pro: An Open-Source Video Editor That Lets AI Agents Drive the Timeline

Most video editors are adding AI features. Palmier Pro is doing something structurally different: turning the editor into an endpoint that AI agents can drive.

Palmier, Inc., a Y Combinator S24 company based in San Francisco, has open-sourced a macOS video editor with a built-in Model Context Protocol (MCP) server. When the app is running, it exposes a local HTTP endpoint at 127.0.0.1:19789/mcp that lets agents like Claude, Codex, and Cursor read and edit the project you have open. No plugins. No copy-paste between windows. 1

The MCP server runs on localhost with origin validation, keeping the connection on your machine. Connecting takes a single command for Claude Code or Codex, a one-click install is available for Cursor, and Claude Desktop support ships bundled with the app as a one-click extension. 2

That architectural choice reframes what a video editor is. The dominant pattern in creative software is AI features bolted onto a human-facing interface: a generate button here, a smart trim there. Palmier Pro inverts the relationship. The editor becomes a programmable surface that an agent can operate on. An agent can cut clips, rearrange the timeline, trigger generation, and work alongside a human editor on the same project, because the project itself is an MCP-accessible endpoint. 1

The editor is built from scratch in Swift 6.2 using SwiftUI, AppKit, and AVFoundation. It targets macOS 26 (Tahoe) on Apple Silicon only. The team says they are building toward a Premiere Pro-caliber editor, with a different take on how AI fits into the workflow. 2

Beyond the MCP server, the feature set reads as deliberate architecture for agent-driven workflows. On-device semantic search uses Google SigLIP 2 converted to Core ML, so an agent can find shots by meaning without sending footage to the cloud. On-device transcription makes spoken content searchable, letting agents navigate footage by what was said rather than scrubbing manually. Export supports MP4 (H.264, H.265, ProRes) or handoff to Premiere Pro and DaVinci Resolve via NLE XML. 1

The generative layer runs five models inside the timeline: Seedance, Kling, Nano Banana Pro, Veo, and Grok. 1

The licensing split is where the strategy becomes visible. The editor, the MCP server, and the in-app chat are all open source under GPL-3.0. The only closed-source component is the generative AI processing layer, which requires a subscription billed with credits. The editor itself is free with no login required. Generative pricing starts at $29 per month for solo creators and $69 per month for heavier use. 2 1

By open-sourcing everything an agent needs to connect and operate on the timeline while keeping the cloud-bound generation layer proprietary, Palmier is making a bet about where ecosystem value lives. Developers who want to build MCP-compatible creative software can study, fork, and extend Palmier Pro's architecture without paying anything. The revenue comes from the compute-intensive generation that no one wants to run locally anyway.

The project drew substantial discussion on Hacker News, where developers praised the decision to build a native Swift app rather than an Electron wrapper and debated whether that choice limits cross-platform reach. One developer building a similar open-source project said they reached the same conclusion about giving agents full access to the app like a real user. 3

Founded by Marcos Rico Peng and Harrison Tin, Palmier ships frequently in the open, with 53 releases to date and v0.3.5 as the latest version. 1

The broader signal for builders is straightforward. MCP is establishing itself as the connective tissue between AI agents and desktop applications. Palmier Pro is a concrete example of what that looks like in creative software: not a chatbot sidebar or a feature toggle, but a full editing environment that agents can treat as a programmable interface. For anyone racing to build MCP-compatible tools, the open-source timeline, the localhost server, and the clean separation between open editor and proprietary generation layer are worth studying.