Setting Up Palmier · Install, Connect an AI Agent, and Edit the Timeline

Quick Summary

• Palmier Pro is a local-first video editor for macOS with a built-in Model Context Protocol server. The editor gives Claude Desktop, Claude Code, Codex, Cursor, and other compatible agents tools for reading a project and making timeline edits.
• In Palmier's setup video, co-founder Marcos installs the app, connects Claude, asks the agent to inspect the open project, uses a transcript to remove filler words and empty gaps, and generates a short B-roll clip that can be placed on the timeline.
• The useful mental model is agent-assisted editing, not a one-click finished film. Let the agent organize a first pass, then review every cut, restore intentional pauses, choose the right takes, refine captions and B-roll, and finish the sound and picture yourself.

What is Palmier Pro?

Palmier Pro combines a conventional multitrack timeline with an MCP server. You can import video, audio, images, and text, then trim, split, reorder, transform, and export clips manually. A connected agent receives a set of tools for understanding the open project and performing many of those timeline operations through natural-language instructions.

That distinction matters. Palmier is the editor and source of project state. Claude, Codex, Cursor, or another MCP-compatible client is the reasoning interface. The agent does not need to guess what is on screen from a screenshot because Palmier can expose structured information about the project, media, transcript, and timeline.

Palmier describes the product as local-first. Regular project files and editing stay on the Mac. The editor can be used without an account, while optional AI generation adds external model calls and credit costs.

Requirements and installation

Start at palmier.io, download Palmier Pro for macOS, and move or open the application as instructed by the installer. Palmier's current documentation requires macOS 26 Tahoe or later. If the app will not launch on an older Mac, check the operating-system version before troubleshooting the agent connection.

On first launch, Palmier presents the project screen and a sample project. The tutorial project is useful because it provides a known timeline for testing the interface and MCP connection before an agent touches client footage.

For a real project, create a project or open an existing one, then import source media into the media panel. You can build the first sequence manually, or connect an agent and ask it to help organize the material after confirming that the correct project is open.

Connect Claude, Codex, Cursor, or another agent through MCP

The video shows an Agent and MCP Server menu in the application menu. Palmier's current documentation now directs users to Help → MCP Instructions. Follow the instructions displayed by your installed version because product menus and connection commands can change.

Palmier currently provides guided or one-click setup for Claude Desktop, Claude Code, Codex, and Cursor. Other MCP-compatible clients can use the server details shown in the same instructions. Complete the installation in the agent you actually plan to use, then return to Palmier with the intended project open.

A connection gives the agent access to Palmier's editing tools. Treat that access like any other local development or automation permission. Read the proposed action, keep backups of valuable projects, and begin with inspection before asking for destructive timeline changes.

Verify the connection before editing

Marcos verifies the setup by asking Claude to read the timeline. This is the right first test because it proves that the agent can see Palmier and the active project without changing the edit.

A useful first prompt is: Read the open Palmier project and summarize its sequence, tracks, media, and transcript. Do not change anything. Compare the response with the visible timeline. If the summary describes the wrong project or misses recent changes, stop and reconnect before continuing.

Next, ask for a proposed edit plan rather than immediate execution. For example: Identify long silent gaps, filler words, and repeated takes. List the cuts you recommend, but do not apply them yet. This separates editorial judgment from tool execution and makes the first agent pass easier to audit.

Build a transcript-aware rough cut

In the tutorial, the agent reads a talking-head transcript, removes empty gaps and filler words, and handles repeated lines by favoring the later take. Marcos says this gets his rough cut roughly 70 to 80 percent complete. That figure is his workflow estimate, not a measured benchmark for every recording or agent.

Transcript editing is most reliable when the instruction defines editorial rules. Tell the agent which filler words may be removed, the minimum silence worth tightening, whether natural breaths should remain, how to choose between repeated takes, and which statements must never be cut. Ask it to preserve sync and report every change.

Review the result from beginning to end. A transcript does not contain every useful pause, facial reaction, visual beat, or continuity issue. Restore pauses that support meaning, check that cuts do not clip consonants, add audio handles or room tone where needed, and confirm that the chosen take matches the picture.

Add B-roll and captions without losing editorial control

Once the dialogue is stable, a connected agent can help identify statements that benefit from B-roll, suggest coverage, create text layers, and place assets at relevant timeline positions. Palmier's MCP tools can also trim, split, reorder, and adjust clips, which makes the same agent useful for the assembly stage.

Ask the agent to propose visual coverage before generating anything. A practical prompt is: Find three moments where B-roll would improve comprehension. For each, quote the transcript line, define the shot, estimate its duration, and explain whether it should replace the speaker or appear as an overlay. Do not generate or place media yet.

Captions need a final visual and editorial pass. Check wording against the audio, break lines at natural phrases, maintain safe margins, and verify readability on a small screen. Generated timing can accelerate the draft, but the finished captions remain part of the publication quality bar.

Generate image, video, and audio inside Palmier

Palmier's media panel can generate images, video, and audio with integrated models. The video demonstrates a six-second cinematic beach clip generated with Grok Imagine, followed by direct placement in the edit. The interface also supports references and first or last frame controls where the selected model allows them.

The available model catalog can change, so choose by capability rather than assuming the exact model shown in the tutorial will always be present. Define aspect ratio, duration, camera behavior, subject action, lighting, and how the shot should connect to the surrounding edit. Generate variations, compare motion across the full clip, then place the strongest result.

Editing and export are currently free, but model generation uses paid credits. Estimate how many variations a shot may require before starting a large batch. External generation also means the supplied prompt and reference media must be suitable for processing by the relevant third-party provider.

Keep the app and MCP tools in sync

Palmier is evolving quickly, and Marcos recommends checking for updates regularly. Use the application's update command, then review the MCP instructions again after installing a new version.

An agent client caches the tools it receives from an MCP server. If Palmier adds or changes tools, an old connection may not expose them correctly. Reinstall or reconnect through Help → MCP Instructions, restart the agent client if requested, and repeat the read-only timeline test.

When a tutorial menu no longer matches your screen, prefer the current in-app instructions and official documentation. Avoid pasting an old server command from a video description into a newer release without checking it first.

Export a finished video or hand it off to another editor

Palmier can export finished video in MP4 formats including H.264, H.265, and ProRes. It can also export NLE XML for continued finishing in Adobe Premiere Pro or DaVinci Resolve.

That makes the agent-assisted rough cut useful even when Palmier is not the final finishing application. You can organize the story and experiment with generated coverage in Palmier, then move the timeline to an established color, audio, graphics, or delivery workflow.

Before handoff, consolidate the sequence, name important media clearly, confirm frame rate and resolution, and open the exported XML in the destination editor. Check transitions, speed changes, generated media, text, and audio routing because interchange formats do not guarantee perfect translation of every effect.

Local-first editing still needs an AI privacy check

Palmier says ordinary projects and media remain on the device unless you invoke an AI feature that requires external processing. When you use generation or the AI assistant, prompts and relevant reference media can be forwarded to the model provider needed for that operation.

Do not upload confidential client material, unreleased products, personal data, or an identifiable person's likeness to an external model without the necessary permission. Read Palmier's current privacy policy and the selected provider's terms before using sensitive inputs.

Generated B-roll and agent-made edits also need human review for factual accuracy, rights, continuity, disclosure, and intent. MCP can shorten the mechanical part of an edit, but the publisher remains responsible for what the final video communicates.

A safe first Palmier workflow

First, install Palmier on a compatible Mac and open the sample project. Second, connect your preferred agent through Help → MCP Instructions. Third, ask the agent to read and summarize the timeline without making changes.

Fourth, import a duplicate or backed-up talking-head project. Ask for a cut plan based on the transcript, approve the rules, and let the agent apply a limited first pass. Review every cut before requesting captions, B-roll suggestions, or generated media.

Finally, refine the sequence manually, check the full export, and use NLE XML if the project needs specialist finishing in Premiere Pro or DaVinci Resolve. The goal is a faster path to an editable first assembly, with the human editor still directing pace, meaning, taste, and delivery.