Edit Client Videos While You Sleep: The Claude + Selects MCP Workflow

TLDR: Most editors are still manually scrubbing footage, connecting Selects to Claude or Codex via MCP creates a fully automated client video editing pipeline that runs while you sleep.

If you're a freelance editor or running a small agency, your margin lives in throughput. Every hour spent on prep work, downloading files, syncing cameras, transcribing, building rough cuts, is an hour you're not billing for something else or taking back for yourself. The workflow in this post eliminates that prep work almost entirely by connecting Claude or Codex to Selects via MCP, creating an automated pipeline that runs without you. Here's exactly how it works.

The Stack: What Each Tool Does

Before walking through the setup, it's worth being precise about what each tool in this workflow actually does, because the pieces connect in a specific order, and confusing them breaks the pipeline.

Claude or Codex via MCP is both the intelligence layer and the coordination layer. When connected to Selects through the Model Context Protocol, Claude or Codex can directly instruct Selects to perform operations on real footage: receiving a client brief, passing footage to Selects, managing the analysis pass, and handling the output. This is the detail that makes the workflow genuinely automated rather than just fast. The AI agent is not generating a text outline for a human to implement. It is orchestrating an actual editing tool on actual footage.

Selects is the editing engine. It ingests the footage the AI agent hands off, runs the full analysis pass, transcription, speaker diarization, silence removal, topic detection, best-take selection, and generates structured draft timelines ready for NLE handoff. Selects does not require an NLE to be open. It works on the footage directly, upstream of Premiere Pro, Final Cut Pro, or DaVinci Resolve.

This two-tool stack is worth noting because most content about AI video editing assumes a three-step workaround: export a transcript, paste it into a chatbot, and manually implement suggestions in your NLE. The MCP connection removes that middleman entirely. As the LLM video editing comparison post covers in detail, Claude and ChatGPT cannot natively process video files through direct LLM interaction. Via MCP, Claude and Codex can instruct Selects to do it for them. The gap is closed.

The Workflow: What Actually Happens

The setup is a one-time configuration. Once it's running, the workflow for each new client project looks like this:

The client sends a Google Drive link. There's no separate upload portal or Dropbox folder, or even back-and-forth about file formats. The client just sends a link, and Claude or Codex handles the rest.

The AI agent downloads the footage and activates the pipeline. The Google Drive link triggers Claude or Codex to pull the raw files and pass them to Selects via the MCP connection. At this point, the editor is not involved.

Selects runs the full analysis pass. Transcription at word-level accuracy, speaker detection from a single audio file, silence removal across all tracks, topic and subtopic detection across the full session, and automatic identification of the best take from repeated attempts at the same line. For a two-hour three-camera podcast recording, this analysis pass takes minutes rather than the four to six hours a manual prep workflow would require.

Selects generates multiple draft versions. The style prompt carries over from the client's previous projects, which means Selects already references the editing style, pacing preferences, and structural patterns that the client expects. This is the detail that makes the workflow feel considered when it runs without human input.

Short-form clips, subtitles, and translations are generated automatically. The pipeline does not stop at the rough cut. Short-form clips are extracted from the draft, subtitles are generated from the transcript, and translation is applied if configured. By the time the editor checks in the next morning, the client has deliverables at every stage of production.

The editor reviews, adjusts, and exports. The human editor's job in this workflow is review and creative refinement, not prep. Draft timelines open directly in Premiere Pro, Final Cut Pro, or DaVinci Resolve via Selects' native NLE handoff. The editorial decisions that require judgment happen here. The hours that used to disappear before the first creative cut are already gone.

Why the MCP Protocol Changes This

The reason this workflow is possible now when it wasn't a year ago is the Model Context Protocol. MCP is a standardised connection layer that lets AI agents like Claude and Codex interact with external tools, in this case, Selects, as if those tools were native capabilities of the AI agent.

Before MCP, using an AI agent for video editing meant copy-pasting transcripts into a chatbot and manually implementing its suggestions in your NLE. The agent could reason about the edit but could not touch the footage. The MCP connection removes that constraint. When Selects is connected to Claude or Codex via MCP, the AI agent can instruct Selects to perform operations on real footage, not describe what those operations should be, but actually trigger them.

This distinction matters for how the workflow scales. An editor managing ten active clients can run ten projects through the same pipeline simultaneously. The AI agent handles the coordination. Selects handles the analysis and assembly. The editor handles the creative decisions. Each layer does exactly what it's best at.

The Style Prompt: The Detail That Makes It Scalable

The most commercially useful feature of this workflow for editors managing multiple clients is the style prompt carryover. Selects references the editing preferences from each client's previous projects and applies them automatically to new ones.

In practice, this means a client who prefers fast-paced cuts with minimal dead air gets that automatically. A client whose content runs longer and prioritises complete answers over tight editing gets that automatically. The AI is not guessing; it is referencing a documented style profile for that client that improves with each project.

For an editor managing recurring clients, this is the difference between scaling and not scaling. Without style memory, editing preferences get re-established on every project. With it, they compound.

Setting It Up: What You Actually Need

The prerequisites are straightforward. A Selects account with MCP integration enabled. Claude or Codex is configured to connect to Selects via the MCP protocol. A Google Drive folder structure that clients can drop footage into. An NLE for the finishing pass. Selects exports natively to Premiere Pro, Final Cut Pro, and DaVinci Resolve.

The video below walks through the full Selects and Claude/Codex configuration and workflow in real time. Watch it first, then use the workflow description above to understand what each step is doing and why.

Who This Is For

This workflow is most valuable for editors billing time to multiple clients on recurring projects. The ROI compounds with volume; the more projects run through the pipeline, the more the style profiles refine, and the less time each project requires in the review stage.

It is also specifically useful for editors who have been limited by the number of projects they can handle simultaneously. The prep work bottleneck is what limits throughput for most freelance editors. Removing it does not just make individual projects faster; it changes how many projects can run at once without adding hours or headcount.

For a deeper look at how Selects fits into a long-form pre-editing workflow beyond the automated pipeline, the complete guide to podcast and interview editing covers the full sequence from raw footage to NLE-ready project.

Try Selects free for 7 days.

Frequently Asked Questions (FAQs)

Q: Can Claude or Codex actually edit video files?
A: Not directly, but via the MCP protocol connected to Selects, they can. When Claude or Codex is connected to Selects through the Model Context Protocol, the AI agent can instruct Selects to perform operations on real footage: transcribing, removing silences, building rough cuts, detecting topics, and generating NLE-ready timelines. The AI agent orchestrates the tool that processes the footage rather than processing the footage itself.

Q: What is the Selects MCP integration?
A: The Selects MCP integration is a Model Context Protocol connection that allows AI agents like Claude and Codex to directly instruct Selects to perform video editing operations on real footage. Rather than a human copy-pasting a transcript into a chatbot and manually implementing suggestions, the AI agent sends instructions to Selects, and Selects acts on the footage directly. The result is a pipeline where the AI agent handles coordination, and Selects handles the analysis and assembly.

Q: Does this workflow work with both Claude and Codex?
A: Yes. The Selects MCP integration works with Claude, Codex, and ChatGPT. The workflow in this post applies regardless of which AI agent you use. If you already have Claude or Codex configured in your workflow, connecting Selects via MCP follows the same process for both.

Q: How much editing time does this workflow actually save?
A: The time saving is concentrated in the pre-editing stage: syncing cameras, transcribing, building rough cuts, and identifying best takes. For a two-hour three-camera recording, this prep typically takes four to six hours manually. Selects completes the same work in minutes via the MCP pipeline. The editor's time in this workflow is spent on creative review and finishing rather than mechanical preparation, which changes how many projects can run simultaneously.

Q: Does Selects remember each client's editing style for future projects?
A: Yes. Style prompts carry over from previous projects, which means Selects references a client's established editing preferences, pacing, structure, and cut tightness, automatically on new projects. For editors managing multiple recurring clients, this is the feature that makes the workflow scalable rather than just fast on individual projects.