documentation

01 overview #

kimo is a free, open-source (MIT) video editor with no GUI, written in Rust. AI agents drive it over the Model Context Protocol (MCP), where every feature is a tool call. It ships as kimo-mcp, an MCP stdio server roughly 15 MB in size with a CPU-only renderer tuned for a 2 GB machine.

runtime footprint #

The kimo-mcp binary runs over standard input and output (stdio) with no background daemon or GPU requirement. Current release is v6.0.0.

system requirements #

kimo requires ffmpeg and ffprobe on your system PATH for decode, encode, and media probing.

02 install #

Choose your operating system below. Your selection is saved locally for future visits.

# installed binary location on windows
$env:LOCALAPPDATA\Programs\kimo\bin\kimo-mcp.exe

mkdir -p ~/.local/share/kimo ~/.local/bin
curl -fsSL https://github.com/naman-sharma12345/kimo/releases/download/v6.0.0/kimo-v6.0.0-linux-x86_64-glibc.tar.gz | tar -xz -C ~/.local/share/kimo --strip-components=1
ln -sf ~/.local/share/kimo/bin/kimo-mcp ~/.local/bin/kimo-mcp

git clone https://github.com/naman-sharma12345/kimo.git && cd kimo && cargo build --release
# binary output: target/release/kimo-mcp

verify ffmpeg and ffprobe #

Confirm that both ffmpeg and ffprobe are reachable on your PATH before connecting an agent:

ffmpeg -version
ffprobe -version

guided setup #

For an interactive step-by-step configurator tailored to your machine and client, visit guided setup → or browse all archives on GitHub Releases.

03 connect your agent #

Point your MCP client at the absolute path to kimo-mcp over stdio.

claude mcp add --transport stdio kimo -- /home/you/.local/bin/kimo-mcp
claude mcp list
# inside a claude code session, run: /mcp

[mcp_servers.kimo]
command = "/home/you/.local/bin/kimo-mcp"

{
  "mcpServers": {
    "kimo": {
      "command": "/home/you/.local/bin/kimo-mcp"
    }
  }
}

{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "kimo": {
      "type": "local",
      "command": ["/home/you/.local/bin/kimo-mcp"],
      "enabled": true
    }
  }
}

{
  "kimo": {
    "command": "/home/you/.local/bin/kimo-mcp"
  }
}

windows path escaping #

04 your first render #

Once kimo-mcp is connected, ask your agent in plain language to import a clip, add a title, and render a frame:

Use kimo to import /path/to/clip.mp4, add a title, and render a frame.

more example prompts #

Three everyday workflows written in the same plain style:

Use kimo to import /path/to/clip.mp4, trim it from 00:00:02 to 00:00:14, and export to /path/to/out.mp4.

Use kimo to import /path/to/clip.mp4, add captions from /path/to/subtitles.srt, and export to /path/to/captioned.mp4.

Use kimo to import /path/to/clip.mp4, color grade it with the LUT at /path/to/look.cube, and export to /path/to/graded.mp4.

05 tool list profiles #

Set KIMO_TOOLS to control how many tool schemas kimo-mcp advertises on startup. Every tool is callable by name under every profile.

setting kimo_tools #

Pass KIMO_TOOLS in your agent's MCP environment configuration:

claude mcp add --transport stdio --env KIMO_TOOLS=core kimo -- /home/you/.local/bin/kimo-mcp

[mcp_servers.kimo.env]
KIMO_TOOLS = "core"

"env": { "KIMO_TOOLS": "core" }

06 how an ai calls tools #

Under the default grouped profile, domain groups accept an action name and an args object, while direct tool names remain callable at any time.

grouped tool calls #

{"name": "vfx3d", "action": "add_scene3d", "args": {...}}

discovery tools #

Agents discover parameters on demand using search_tools(query, limit=8, group?) to find matching operations and kimo_describe to inspect full input schemas before invoking a tool.

// search by keyword and optional group, then inspect schema
search_tools(query, limit=8, group?)
kimo_describe

browse every tool →

07 environment variables #

Environment variables read by kimo-mcp:

variable reference table #

08 troubleshooting #

kimo doesn't appear in my agent #

Use an absolute path to the kimo-mcp binary (do not rely on ~ expansion inside JSON configs), run chmod +x on macOS/Linux, and restart the agent.

ffmpeg not found #

Install ffmpeg (which includes ffprobe), make sure both are on your system PATH, and open a new terminal before launching your agent so it inherits the updated environment.

smartscreen blocked the installer #

The Windows installer (kimo-6.0.0-windows-x86_64-setup.exe) is not code-signed yet. When Windows SmartScreen prompts you, click More info and then Run anyway.

my agent lists too many tools #

Set KIMO_TOOLS=grouped (the default, 17 entries and 4,467 bytes) in your MCP client config so your agent uses grouped calls and on-demand discovery.

09 contributing #

kimo is MIT-licensed and developed in the open at github.com/naman-sharma12345/kimo. Read the architecture and extension guides before opening a pull request:

repository guides #