Install the agent-share CLI, sign in with your Agentics account, and wire your repo so a coding agent can read your phone shares and act on its inbox.
agent-share login && agent-share initGet a coding agent connected to your Agent Share inbox in a few minutes: install the CLI, sign in, wire your repo, and confirm a phone screenshot reaches the agent.
This guide gets you to a working setup. For full command and tool details, see the CLI reference.
Before you start, make sure you have:
The shipped product is a pull inbox: your agent reads new items when it calls share.list. Push-into-session channels are a Claude Code research-preview feature that needs Claude Code v2.1.80+ and extra wiring — see Channels & the agent inbox. You do not need channels for this quickstart.
Pick one. The one-line installer is the recommended path; npm and a manual binary download are alternatives.
The installer detects your OS/architecture, verifies the checksum, drops the binary on your PATH, and always fetches the latest published release.
Linux / macOS
curl -fsSL https://agentics.dk/install/agent-share.sh | bash
Windows (PowerShell)
irm https://agentics.dk/install/agent-share.ps1 | iex
The npm package downloads the platform binary for you. Available since 1.7.0.
npm i -g @agentics-live/agent-share
Release binaries for all platforms ship with each release (since 1.6.0). Download the build for your OS/arch from the project releases page and put it on your PATH.
agent-share version
You should see agent-share <version> printed.
Run the login command. It opens your browser and runs an OIDC loopback flow against login.agentics.dk/realms/agentics.
agent-share login
On success the CLI stores a refresh token (plus your sub and display name) in ~/.agent-share, then sanity-checks it with a GET /api/agents call.
No token is written into any project file — every command auths from this stored login and refreshes silently. If the saved refresh token ever goes dead, the CLI re-runs login inline on a TTY automatically (since 1.15.0), or prints copy-paste re-login instructions when running non-interactively.
Choose how the repo connects to Agent Share. All three commands write a project-scoped ./.mcp.json and drop a SKILL.md. Full detail lives in the CLI reference and the Claude Code guide.
agent-share init — your personal inbox. Best when you want the agent to read your own phone shares and share.send back to buzz your phone. Wires ./.mcp.json to agent-share mcp --personal. No token in the file (auth = your stored login).
agent-share init
agent-share install — agent pull tools. Wires agent-share mcp (the pull-style agent tools) into ./.mcp.json. Variants: install channel for channel mode, install aspire for a local Aspire dev server.
agent-share install
agent-share register "name" — a named inbox. Enrolls a named inbox, wires agent-share channel pinned to it (server + token live in the entry's env), and prints the A2A card plus the channel launch line.
agent-share register "my coding session"
If you are unsure, start with agent-share init.
First, confirm your inbox is registered and online:
agent-share agents
You should see your inbox listed with an online status dot.
Then test the full path:
share.list — the new item should appear.Reading does not mark an item; the agent acknowledges it explicitly with share.mark_received. By default share.list returns only new (unacknowledged) to_agent items.