Command line and AI agent access (CLI, MCP, agent skill)

Summary

Drive Snim from a terminal, a script or an AI coding agent: take a screenshot, record the screen, read text with OCR and upload a file, without touching the tray. Nothing is reachable until you switch it on in Settings.

What you will be able to do

• Run commands such as snim capture, snim record start, snim record stop, snim ocr and snim upload, for a whole screen, an area you give in pixels, or a file.
• Add --json to any command to get results a script can read, such as the saved file's path or the uploaded link.
• Record for a fixed time, for example snim record start --duration 20, and get the finished file back.
• Let AI agents such as Claude Code, Codex and Cursor use Snim through its built-in MCP server (snim mcp), so they can take a screenshot and look at it while they work.
• Set up an agent with one command, snim agents install, or a button in Settings: it adds a Snim skill and connects the MCP server.
• List screens and windows with snim targets, and check access and permissions with snim doctor.

How it works

• The snim command talks to the running Snim app, and starts it in the background when needed. Captures and recordings happen inside the app, so the permissions you gave Snim keep applying.
• Access is off by default. Turn on "Allow command-line and agent access" in Settings > Agents, where a short log shows recent commands.
• When a command captures or records, you can see it: the tray icon changes, recordings show the usual controls, and a notification appears when a recording starts.
• Upload passwords never leave Snim; a command only gets the link back.

Platforms

• Windows: snim works in Command Prompt and PowerShell, and the installer can add it to your PATH.
• macOS: The Homebrew cask adds the snim command; with the DMG, install it from Settings. Screen Recording permission still belongs to Snim, not your terminal.
• Linux: The packages and the install script already provide snim; in the Flatpak, run flatpak run dev.snim.Snim followed by the command, and files are saved in Pictures/Screenshots and Videos/Snim. Captures are silent on KDE Plasma, wlroots desktops and X11, and on GNOME after you choose your screens once. On Wayland, recordings will remember the chosen screen after the first pick, and window lists are not available, so agents work with screens and areas.

Not in this version

• Local use only: commands work for your own user on the same computer, never over the network.
• No per-command approval prompt yet: access is on or off.
• Window captures cut the window's area out of the screen, so anything overlapping it shows too.

Status

• Command channel, --json, doctor, targets and the Settings switch
• Screenshots and OCR from the command line
• Uploads that return the link
• Recording start, stop and fixed-length recordings
• MCP server for AI agents
• Agent skill and one-step setup for Claude Code, Codex and Cursor
• snim on the PATH everywhere, and a setup guide on snim.dev