Install · Agentry docs

The plain install

Agentry installs as a Claude Code plugin from its GitHub repo. From a Claude Code session:

$ /plugin marketplace add Codestz/agentry
$ /plugin install agentry@agentry-dev

That's the whole setup. The conducting front door (/agentry:go), the SDLC specialists, and the durable memory layer all run from here — entirely from the terminal, no account and no build step. Head to Quickstart to hand it a first task.

Requires Node ≥ 24 — the memory engine uses Node's built-in node:sqlite, so there are no native modules to compile.

Bringing channels online

Channels add live human↔agent push — comments, threaded replies, a permission relay, and status steering — on top of the install above. They are an additive layer and a research preview. How you launch depends on which channel you're enabling, because every channel must clear Anthropic's approved allowlist to register.

Official channels — the normal way

A channel on the approved allowlist — the official set in claude-plugins-official (Telegram, Discord, iMessage) — is enabled per session with the --channels flag. No bypass, no warning; the channel is already trusted:

$ claude --channels plugin:telegram@claude-plugins-official

This is the path a normal user takes for an officially-supported channel.

Agentry's channel bridge — the development flag

Agentry's own channel bridge is a custom, research-preview channel. It is not yet on the approved allowlist, so --channels won't load it. You bring it up with the development-channels flag, which bypasses the allowlist for that one entry after a one-time confirmation prompt:

$ claude --dangerously-load-development-channels plugin:agentry@agentry-dev

Why two flags --channels enables a channel that is already on Anthropic's curated allowlist; --dangerously-load-development-channels bypasses that allowlist per-entry, so you can run a channel you're developing — or one not yet officially listed, like Agentry's bridge today. The bypass is per-entry and skips the allowlist only; an organization's channelsEnabled policy still applies, and you shouldn't point it at channels from untrusted sources.

Why the flag is load-bearing — and what happens without it

For Agentry's bridge the flag is load-bearing in a subtle way: the channel's MCP instructions load without it, so seeing them is not proof the channel is live. The flag is what actually brings the channel up — it is the thing that proves the channel is running, not just described.

Leaving the flag off is a graceful degradation, not a breakage. Everything still works via the async file-watch loop: the files under .agentry/work/ are the source of truth, the agent reads your edits on its next turn, and tool-use approvals fall back to the terminal. You only need the flag when you want the live push instead of the next-turn loop.

Preview The --dangerously-load-development-channels flag is a research preview — copy it verbatim, and expect it to change before 1.0.

Next: Channels for what live push gives you, and Permissions for how the relay approves a tool call from the Workbench.