Docs / Reference

The tezgah command

Control Tezgah from any of its terminals: notifications, status, panes, sessions, the browser, and an MCP server for agents.

Every terminal inside Tezgah has a tezgah command. It needs nothing installed: it runs on Tezgah’s own runtime and talks to the app over a socket only your user can open. Inside Tezgah, this pane is the terminal you run the command in.

tezgah help lists everything.

Sessions and panes

Command What it does
tezgah notify [--title <title>] <message> Notify: rings the pane and shows in the panel
tezgah status <working|done|blocked|idle|clear> [message] Set this pane’s status in the sidebar
tezgah split [right|down] [-- <command>] Split this pane; run a command in the new one
tezgah new [--shell] [--dir <path>] [--title <title>] [-- <command>] New session (Claude by default)
tezgah open <url|folder> A URL opens in a browser pane, a folder as a project
tezgah send [--pane <id>] [--enter] <text> Type text into a pane (default: this session’s Claude)
tezgah read [--pane <id>] [--lines <n>] Print what a pane shows (default: this one)
tezgah list [--json] Sessions and panes
tezgah run <name> Run one of your custom commands
tezgah focus <session-id> Show a session

Browser

These act on this session’s browser pane; --pane <id> picks another.

Command What it does
tezgah browser open <url> [--new] Open a page (in a new pane with --new) and wait for it
tezgah browser snapshot [--interactive] The page as text, every element with a ref (e12)
tezgah browser click <target> [--double] [--right] Click
tezgah browser fill <target> <text> Replace a field’s text
tezgah browser type [--into <target>] <text> Type at the focus, or into a field
tezgah browser press <keys…> Enter, Tab, Escape, ArrowDown, Control+a…
tezgah browser select <target> <option> Pick an option in a <select>
tezgah browser hover <target> Hover
tezgah browser scroll [up|down|top|bottom] [--to <target>] Scroll
tezgah browser screenshot [<target>] Save a PNG and print its path
tezgah browser wait [--text <t>] [--gone <t>] [--selector <css>] [--url <part>] [--ms <n>] [--timeout <ms>] Wait for something
tezgah browser console [--errors] [--clear] What the page logged
tezgah browser network [--failed] [--clear] Requests it made
tezgah browser text | url | eval <js> | back | forward | reload Read the text or URL, run JavaScript, navigate

A target is a ref from the latest snapshot (e12), a CSS selector, or text=<visible text>.

Examples

# Run the dev server beside you, then check the page
tezgah split down -- npm run dev
tezgah browser open http://localhost:5173
tezgah browser screenshot

# Tell yourself when a long job is done
make release && tezgah notify --title Release "built"

# Start a shell session in another project, running a command
tezgah new --shell --dir ~/code/api -- make test

# Ask this session's Claude to try again
tezgah send --enter "run the tests again"

MCP

tezgah mcp runs an MCP server over stdio with the same tools as the browser tools. Claude sessions get it automatically; add it to other agents as shown in The browser.

Something missing? Open an issue