SkillAgentSearch skills...

Montagent

Agentic, MCP-native video editor for the command line. Cut, trim, and compose video through CLI commands or AI agent tool calls.

Install / Use

claude mcp add MBehtemam -- npx -y github:MBehtemam/Montagent

If the server publishes to npm under a different name, use that package instead — check the repo README.

About this skill
🔌

MCP Server

Model Context Protocol server

Quality Score

81/100

Supported Platforms

Claude Code
Claude Desktop

Our assessment of Montagent

Montagent scores 81/100 on our quality scale, 3483rd of 4,650 Development & Engineering skills we index.

Its MCP Server is 6.7 KB long, well organised into 11 sections with 5 code examples: a thorough specification that gives an agent plenty to work with.

It has 3 GitHub stars, so there is little community track record yet; judge it on its content.

Substance
29/30
Structure
20/20
Description
15/15
Adoption
3/20
Freshness
15/15

Maintenance, license and trust

  • The repository was last updated 3 days ago, so Montagent is actively maintained.
  • No license is declared. By default that means all rights are reserved: you can read it, but reusing or redistributing it is not clearly permitted. Ask the author before building on it commercially.
  • Its trust signals score 80/100, with 2 cautions from licensing, adoption, age or documentation. These come from repository metadata, not a code audit — read the skill file before letting an agent act on it.

Montagent compared with similar skills

All 4 of these similar skills score higher than Montagent; compare them before choosing.

SkillScoreStarsUpdatedFormat
Montagent (this skill)by MBehtemam8133d agoMCP Server
Agent-Reachby Panniantong100106.0k4d agoCLAUDE.md
taste-skillby Leonxlnx100105.1ktodayCLAUDE.md
headroomby headroomlabs-ai10078.2ktodayCLAUDE.md
ai-job-searchby ai-job-search-dev10067.4k2d agoCLAUDE.md

Frequently asked questions

How do I install Montagent?
Run claude mcp add MBehtemam -- npx -y github:MBehtemam/Montagent. The install tabs above show the steps for each supported agent.
Which AI agents does Montagent work with?
It is written for Claude Code and Claude Desktop, as a MCP Server file. Other agents that read the same format can often use it too.
Is Montagent safe to use?
It declares no license and scores 80/100 on trust signals. Skills are instructions an agent will follow, so read the file before installing it and do not approve commands you do not understand.
Is Montagent still maintained?
The repository was last updated 3 days ago, so Montagent is actively maintained.

Montagent

Montagent is a video editor whose project format is designed to be authored and edited by an AI agent rather than dragged around in a GUI. You describe what should be on screen to an agent — Claude Code, or any other client speaking MCP, the protocol that lets an AI assistant call a tool — and Montagent renders it.

Why not a timeline GUI?

A GUI timeline is built for a hand on a mouse: drag a clip, nudge a keyframe, look at the result. An agent doesn't have a hand or an eye — it has a document. A Montagent project is a single JSON file that states, by being read, what is on screen at any given moment: which track, which element, what time range, what it looks like. An agent edits that file the way it edits code — read, change a value, re-render, look at the frame — and the file is the whole authority: check it into git, diff it, review it in a pull request. There's no separate project database, and nothing "in the app" that a diff can't show.

Montagent itself is deterministic and contains no model: it doesn't write your video for you, it renders the one you (or your agent) described, the same way every time. It's also fast — a 2-second, 1920×1080 clip typically renders well under half a second.

Install

cargo install montagent

or download a prebuilt binary for your platform from the Releases page and put it on your PATH. Six desktop targets are built: macOS, Linux and Windows, each for x86_64 and aarch64. CI runs the test suite on five of them. Intel macOS (x86_64-apple-darwin) is still built and released, and CI checks that it builds, but its tests are not run (ADR-0188).

You'll also need ffmpeg on PATH: 7.1 or newer, built with libx264 (ADR-0115). Ubuntu 24.04's apt ships 6.1, which is below that; use a static build or a newer distribution. Montagent spawns it for encoding and decoding rather than linking it in (its codecs are GPL; Montagent's binary isn't), and it's only asked for when you actually render — validating or scaffolding a project needs nothing but the montagent binary itself.

Linux, containers and hosting

The Linux binaries run in a container or on a server today: install ffmpeg 7.1 or newer and run montagent as above. A hosted, remote MCP server (an HTTP door and named workspaces, ADR-0064, ADR-0159) is designed but not shipped: its container image is tracked in #656, and long renders over a remote connection in #662.

Same project, same bytes: per target

A project renders to the same frames and the same MP4 bytes every time on one target (one platform and architecture, one ffmpeg build), whatever the painter count or chunk size (ADR-0143). Across architectures there is no promise: x86_64 and aarch64 may differ, and there is no tolerance that makes them count as equal. On x86_64 Windows the guarantee is the frames handed to the encoder, not the file's bytes (ADR-0190).

macOS: signed and notarized after v0.1.0

Release binaries after v0.1.0 are signed with a Developer ID and notarized by Apple (ADR-0138), so the binary in the tarball runs as extracted — including when an MCP client starts it. macOS checks it with Apple the first time it runs, which needs an internet connection.

v0.1.0 is unsigned. macOS quarantines its extracted binary and kills it silently — not with a dialog you can click through, but a kill on the first attempt and, after that, a kill with no dialog at all. The same fix also covers a signed binary whose first run has to happen offline. Before running it the first time:

xattr -d com.apple.quarantine ./montagent

No sudo, no GUI, and nothing to click. cargo install never hits this — cargo builds the binary on your machine rather than downloading a quarantined one.

Don't use spctl --assess to check the binary: it reports rejected for every command-line tool, signed and notarized or not.

Wire it up to an agent

Montagent is a single binary that is both this CLI and an MCP server. Point your MCP client at it with a config block like this one (Claude Code / Claude Desktop's shape — adjust the wrapper for whatever client you use):

{
  "mcpServers": {
    "montagent": {
      "command": "montagent",
      "args": ["mcp"]
    }
  }
}

Once it's connected, your agent has thirteen tools — compare, create_project, fonts_acquire, fonts_list, fonts_vendor, frame, measure, preview, query, render, shift, validate, verify — and resources that teach it the project format directly: montagent://format.md with the pages it lists, and the schema, whole at montagent://schema.json and in pieces listed by montagent://schema/index.json. That's where the agent learns the verbs; this page isn't trying to repeat them.

See it work, without wiring up an agent

Unpack the release archive (or clone the repo) and run, from its root:

montagent fonts vendor examples/hello-text/hello-text.montagent.json examples/hello-text/OpenRunde-Bold.otf --licence OFL-1.1
montagent render examples/hello-text/hello-text.montagent.json

That writes out/hello-text.mp4 — 90 frames of white text on a dark background — and only the second command needs ffmpeg.

The first command isn't a formality: Montagent refuses to render a font it hasn't verified the licence of, even the Open Runde font shipped alongside this example, so --licence OFL-1.1 is you confirming what OpenRunde-LICENSE.txt (also in that directory) already says. See examples/hello-text/README.md for why.

If a source looks stale

Montagent remembers what it probed about each media file in a per-user cache, so an unchanged file costs no ffprobe. It notices a file that changed on disk, including one rewritten in place under an unchanged size and timestamp. If you ever need to start from the disk again:

montagent cache clear

It prints the path it removed, which is platform-specific and not worth memorising.

Learn more

CONTEXT.md and docs/adr/ hold the reasoning behind the format and every non-obvious decision in it, for anyone who wants more than this page says.

Licence

MIT — see LICENSE. The test fixtures under fixtures/ and some committed research media are not covered by it; see LICENSE-MEDIA.md for the carve-out. The binary itself statically links Skia and around 128 other Rust crates; THIRD-PARTY.md — shipped in this repo and in every release archive — lists what's owed for them.

Explicit font acquisition, portable licence evidence, and CLI/MCP examples are documented in Font acquisition.

Related Skills

View on GitHub
GitHub Stars3
CategoryDevelopment
Updated3d ago
Forks0

Languages

Rust

Trust signals

80/100

From repository metadata: license, adoption, age and documentation. Not a code audit — see the Safety scan above for what the skill file itself contains.

1 medium1 low