Edit videosLarge Language Models

MCP Server for Video Editing: Free Options, Claude Setup and Tools That Work

Which MCP server for video editing is actually free, how to connect it to Claude on your own machine, and which tools trim, cut, caption and render footage reliably. Real limits, costs and safe settings included.

MCP Server for Video Editing: Free Options, Claude Setup and Tools That Work
Cristian Da Conceicao

You have forty minutes of raw footage, a client who wants three vertical clips by tonight, and a timeline that still needs a first pass. Now picture typing "cut the silences, trim the intro and export a 9:16 version" and watching your files change on your own computer. That is the promise of an MCP server for video editing, and in late 2026 it is real enough to use, as long as you pick the right server and keep your expectations grounded.

This article sorts the free options from the paid ones, shows how to connect a server to Claude, and tells you honestly where each tool works and where it falls over.

What an MCP Server Actually Does

A developer typing in a terminal at a wooden kitchen table in morning light

The short version

MCP stands for Model Context Protocol, an open standard that lets an AI assistant call tools on your machine or on a remote service. An MCP server is a small program that exposes a list of actions, such as "trim this clip", "concatenate these files" or "add a text overlay". The assistant reads that list, decides which action fits your request, fills in the arguments, and the server runs it.

The model never touches your video directly. It sends instructions, and the server does the work. That separation is why the setup is both powerful and predictable: the tools are fixed, so the assistant can only do what the server allows.

Why video editing fits so well

Most editing chores are repetitive and describable in a sentence. Trim the first eight seconds. Convert everything to H.264. Burn subtitles into the lower third. Resize to square. These are exactly the tasks where a plain-language request beats clicking through menus, and where a script would otherwise take twenty minutes to write.

What MCP does not do well is taste. It will not decide that a cut should land half a beat late or that a reaction shot deserves two more frames. Think of it as a very fast assistant editor for the mechanical parts of the job.

💡 Rule of thumb: if you can describe the edit as a command, an MCP server can probably do it. If you describe it as a feeling, you still need your own eyes.

Free Options That Run Locally

A neat bundle of cables plugged into a compact home server and network switch

FFmpeg-based servers

The clearest free path is a server that wraps FFmpeg, the open-source engine behind most of the video software you already use. A few names turned up repeatedly in recent roundups:

  • mcp-video: an open-source Python server that wraps FFmpeg and Remotion, listed with roughly 82 tools for AI agents. It runs locally and costs nothing.
  • ffmpeg-skill and Kinocut: free servers for processing local files, with ffmpeg-skill listed under an MIT license.
  • free-video-edit-mcp: an MIT-licensed option for FFmpeg and ffprobe editing, and it states clearly that it has no connection to Blackmagic or DaVinci Resolve.

Because all of these call FFmpeg underneath, the feature set is similar: trim, merge, resize, convert, extract audio, change speed, add text, burn subtitles, generate thumbnails. There is no timeline you can open afterward, which matters if you want to keep editing by hand. You get finished files, not a project.

💡 Star counts and "free" labels on directory sites come from third parties. Open the project's repository, check the license file and read the last commit date before you install anything.

What free really costs

The software is free. Your time is not. Expect to spend an hour on installation the first time, mostly getting FFmpeg onto your system path and a Python or Node runtime working. Expect to debug at least one quoting problem with spaces in file names. And expect that the assistant occasionally picks a clumsy FFmpeg command that works but re-encodes more than it needs to.

An aerial view of a student's desk with a budget laptop showing a simple video timeline

Here is how the free path compares with the paid editors that have bridges:

OptionCostOutputNeeds paid appBest for
FFmpeg-based serverFreeRendered filesNoBatch trims, conversions, captions
DaVinci Resolve bridgeFree server, paid Studio for external scriptingReal project and timelineYes, Resolve StudioEdit and color work you will continue by hand
Premiere Pro bridgeFree server, paid subscriptionPremiere projectYesTeams already living in Premiere
Hosted video MCP serviceUsually a subscription or creditsCloud rendersNoPeople who do not want local installs

Resolve and Premiere Bridges

DaVinci Resolve

The most starred community project in this space connects Claude to DaVinci Resolve through its official scripting interface, and it is open source under an MIT license. It can create projects, build timelines, add clips and manage media pools, which means the output is something you can open and refine.

The catch is licensing. External scripting requires DaVinci Resolve Studio, the paid version. The free edition of Resolve can only reach a server through an in-app bridge script, which is more fiddly and less stable. Another project can generate an OpenTimelineIO file that you import into Resolve Studio, a useful middle road if you want the assistant to plan an edit without driving the app live.

A colourist's hands on a grading control surface in a dim suite

Premiere Pro

There is no official Adobe MCP server. What exists are community bridges, a couple of them MIT licensed, that talk to Premiere through a panel running inside the app. You need a Premiere subscription, and because Adobe changes its scripting surface from time to time, a bridge can break after an update.

I would treat the Premiere route as a convenience for people who already pay for the app and have a specific repetitive job to hand off, like creating sequences from a folder of clips. It is not a reason to subscribe.

Two editors at adjacent desks with large monitors showing multi-layer timelines

Claude Setup Step by Step

A woman configuring a laptop with indented configuration text on a dark editor

Before you start, install the basics: FFmpeg (confirm that ffmpeg -version works in a fresh terminal), plus the runtime your chosen server needs, usually Python with a package runner or Node. Then pick one of the two paths below.

Claude Desktop

Claude Desktop reads a JSON configuration file where you list your servers. Open the app's settings, find the developer section, and choose the option to edit the config file. Add an entry under mcpServers:

{
  "mcpServers": {
    "video-editor": {
      "command": "uvx",
      "args": ["mcp-video"]
    }
  }
}

The exact command and arguments depend on the server, so copy them from its README rather than from this example. Save the file, fully quit and reopen the app, and look for the tools icon in the chat box. If your server appears with its tool list, the connection works.

Claude Code

If you work in a terminal, Claude Code adds a server with a single command. You give the server a name, then the command that launches it. Run claude mcp list afterward to confirm it registered. This route suits developers who want to keep video chores next to scripts and folders, and it makes it easy to scope a server to one project directory.

Test with a tiny task

Do not start with your real footage. Copy a ten-second clip to a scratch folder and ask for something verifiable:

  1. "List the duration, resolution and frame rate of this clip."
  2. "Trim it to seconds two through six and save as a new file."
  3. "Resize the trimmed clip to 1080 by 1920 with a blurred background."

If step one works, the server can read files. If step two works, it can write them. Step three tells you whether the assistant chooses sensible FFmpeg options. Only then point it at real material.

💡 If a tool call fails with "file not found", the cause is almost always a relative path or a space in a folder name. Give absolute paths and keep scratch folders simple.

Model choice matters too. A capable coding-oriented model plans multi-step edits more reliably than a small, fast one. Anthropic's Claude Sonnet 5 and Claude Fable 5 are the sort of models people reach for when a job involves chaining several tool calls with the right arguments.

Tools That Work in Practice

A creator holding a smartphone showing a vertical clip with a podcast microphone nearby

Batch social clips

This is the strongest use case. Give the assistant a folder of long recordings and a list of timestamps, and ask for vertical clips with captions burned in. FFmpeg-based servers handle this well because every step is deterministic: cut, crop, scale, overlay text, encode. A folder of twenty clips that would take an afternoon of clicking becomes a few minutes of waiting.

Tips that save frustration:

  • Ask for a filename pattern up front, so outputs do not overwrite each other.
  • Request a dry run first. Have the assistant describe the commands it plans to use before it runs them.
  • Keep source files read-only and write results to a separate output folder.

Rough cuts from long footage

Another reliable job is a first-pass assembly. The assistant can detect silence, list scene changes and propose a cut list; you review it and approve. With a Resolve bridge the result lands on a real timeline. With an FFmpeg server you get concatenated files that you can drop into any editor.

Where it struggles: anything that needs visual judgment. Choosing the best take, matching action across cuts, or spotting a boom mic in frame is still your job. Some servers add frame sampling so the model can look at stills, but treat that as an assist, not a verdict.

TaskReliabilityNotes
Trim, merge, convertHighDeterministic FFmpeg commands
Burned-in captionsHighNeeds a transcript or subtitle file
Silence removalMedium to highCheck cuts near breaths and laughs
Reframing to verticalMediumSubject tracking is limited, so verify crops
Timeline building in ResolveMediumDepends on Studio and version match
Creative pacing decisionsLowKeep a human in charge

Safety and Permissions

A hand sliding a silver external SSD into a laptop beside a brass padlock

Keep files scoped

A local server runs commands on your computer. One roundup of FFmpeg servers pointed out exactly that: they execute directly on your machine. Treat that seriously. Install servers only from repositories you have read, prefer projects with recent activity and a clear license, and avoid anything that asks for far more access than editing needs.

Limit the working directory. If the server supports an allowed-paths setting, point it at a single project folder instead of your whole drive. If it does not, run it from a user account or a container that only sees your footage.

Review before you render

Most mistakes are cheap if you catch them early and expensive if you catch them after a four-hour batch. Ask the assistant to print the command list and read through it. Look for overwrite flags, odd output paths and deletes. Back up originals to an external drive before the first real run.

💡 Never paste access tokens or passwords into a chat so a server can "log in" for you. If a hosted service needs credentials, store them in its own settings.

Where PicassoIA Editing Models Fit

A filmmaker reviewing footage on a laptop on a rooftop terrace at golden hour

An MCP server handles the mechanical side of editing: cutting, converting, captioning, assembling. It does not repaint a scene, swap a jacket colour, or remove a stray person from the background. That kind of generative edit is a different job, and it belongs to dedicated video models. If you want one workflow with both halves, use an MCP server for the file handling and an AI editing model for the creative changes.

None of the models below is an MCP server, and this is not a claim that they plug into Claude. They are simply the browser-based tools I would pair with an FFmpeg pipeline.

Text-driven edits

  • P Video Edit changes a video from a written prompt, handy for quick fixes to lighting, wardrobe or setting.
  • Wan 2.7 Videoedit and Lucy Edit 2 both edit footage from text, so you can compare how each reads the same instruction.
  • Kling o1 rewrites a clip with text and suits restyling a shot while keeping its motion.
  • Aleph 2 edits a single frame and carries that change through the whole video.

A sensible order of work: use your MCP server to trim a long recording into a clean ten-second clip first, then send only that short clip to an editing model. Shorter inputs finish faster and keep your results easier to compare.

Audio and framing

  • Video Erase Object removes a distracting object from footage, something FFmpeg cannot do.
  • Reframe Video changes the aspect ratio of a clip, a good companion to batch vertical exports when automatic crops cut off the subject.
  • Video To SFX v1 adds synced sound effects, which helps when your assistant has trimmed a clip that now feels silent.

Common Mistakes to Avoid

Most failures I see come from the same handful of habits:

  1. Starting with real footage. Test on a scratch copy first.
  2. Vague requests. "Make it better" gives the model nothing to execute. Name the length, the ratio, the codec or the caption style.
  3. Installing five servers at once. Each one adds tools to the list, and the assistant gets worse at choosing between near-duplicates. Start with one.
  4. Ignoring version mismatches. A Resolve or Premiere bridge built for one app version may misbehave on another. Check the README for supported versions.
  5. Skipping a backup. Keep untouched originals somewhere the server cannot write.

Try It Yourself on Picasso IA

If the setup above feels like a lot, start with the creative half and skip the install. Open Picasso IA, upload a short clip, and test one of the video editing models with a single clear instruction: change the time of day, swap a background, or erase one object. Run the same prompt on two models and compare the results side by side.

Then, when you are ready to automate the repetitive parts, install one free FFmpeg-based MCP server, connect it to Claude with a ten-second test clip, and build from there. A small, working pipeline beats an ambitious one that never leaves the config file. Pick a clip you already have, try a first edit today, and see how much of your routine it can take off your plate.

Share this article

Choose your language

FreeCreate AI Images and VideosFree, No Credit CardFree Generations to StartSeedance 2.5 Lite IncludedStart Free Now