What is the difference between an MCP server and an Agent Skill?

A skill is code that runs where the agent runs, on your own machine and with your own toolchain. An MCP server is a service the agent calls over a protocol, so the work happens somewhere else. That one difference decides whether you can use it from a phone, call it from a scheduled job, or share it across a team.

Is Playhead open source?

The MCP client and the TypeScript SDK are MIT-licensed and on npm. The engine is a hosted service. claude-video and claude-watch are MIT throughout, which is a genuine advantage of theirs and is listed as one.

Do the alternatives need an OpenAI or Groq key?

Both use the uploader's captions when a video has them, and fall back to Whisper through a Groq or OpenAI key you supply when it does not. Playhead includes transcription in a credit, so there is no second vendor.

Which should I use for footage under NDA?

A local skill, or your own script. Both process entirely on your machine, and no hosted service can match that. Playhead's on-premise deployment exists for enterprise, but the hosted product downloads and processes on our infrastructure.

Comparisons

The alternatives to Playhead are Agent Skills: Python scripts that run on your own machine and need a video toolchain and usually a transcription key of your own. They are free and they work well on a laptop. What they cannot be is a connector. A skill that needs a toolchain cannot run from an agent on a phone, and your production code cannot call it either.

Last updated 31 August 2026

The alternatives

Three real ones. Two are MIT-licensed projects by people solving the same problem, and both are good at what they were built for.

The one difference that decides it

Not features. Where the work happens, because everything else follows from that.

claude-video and claude-watch are Agent Skills: Python that runs where the agent runs, shelling out to yt-dlp and ffmpeg on your machine, with a Groq or OpenAI key of yours for anything without captions. They are free, they are MIT, and on a laptop they work well.

Playhead is a hosted MCP server. The toolchain is ours to keep working, and the consequences are the whole comparison:

What follows from where the work happens
Because the work happens…	…this becomes possible	…and this is the cost
on our servers	It runs from Claude on a phone, where there is nowhere for ffmpeg to live.	Your video leaves your machine.
behind an HTTP API	A backend job can call it at three in the morning with no agent involved.	There is a bill.
against one account	A team shares a balance, a history and an audit trail.	You depend on a vendor staying up.
in one place	Nine platforms keep working when they change their minds about who may download what.	You cannot patch it yourself.

There is a second difference, and the tables are mostly about that one: depth. A skill hands the model frames and lets it work out the rest. Playhead spends the same download on a second pass that reads every frame, so a cut list, named transitions, camera moves with directions, overlay spans with positions and a statement of which seconds were never looked at come back alongside the pictures. On a fifteen-second ad the two approaches land close together. On four minutes they do not.

How we compare

Every claim about somebody else's project on these pages comes from their own README, read on 6 August 2026. Where their documentation is silent, the table says so rather than filling the gap with the answer that flatters us.

That is not politeness. A comparison table is the easiest page to get caught lying on. The other project's maintainers read it, and one overstated row discredits the twenty accurate ones beside it. So Not documented is a real value in these tables, and it is used.

Each page also lists what the other project is better at, in its own section, before the table. A reader who discovers those elsewhere discovers them along with the fact that we left them out.

Questions

What is the difference between an MCP server and an Agent Skill?: A skill is code that runs where the agent runs, on your own machine and with your own toolchain. An MCP server is a service the agent calls over a protocol, so the work happens somewhere else. That one difference decides whether you can use it from a phone, call it from a scheduled job, or share it across a team.
Is Playhead open source?: The MCP client and the TypeScript SDK are MIT-licensed and on npm. The engine is a hosted service. claude-video and claude-watch are MIT throughout, which is a genuine advantage of theirs and is listed as one.
Do the alternatives need an OpenAI or Groq key?: Both use the uploader's captions when a video has them, and fall back to Whisper through a Groq or OpenAI key you supply when it does not. Playhead includes transcription in a credit, so there is no second vendor.
Which should I use for footage under NDA?: A local skill, or your own script. Both process entirely on your machine, and no hosted service can match that. Playhead's on-premise deployment exists for enterprise, but the hosted product downloads and processes on our infrastructure.