Claude Code

In Claude Code, the Cranes plugin makes Cranes automatic. Without it, Claude Code decides for itself when to call Cranes.

What’s automatic with the plugin

• Your thinking on every prompt. Before Claude Code works on a message, the plugin asks Cranes how your thinking bears on it and on the work, and adds what comes back for Claude to use. Type @cranes any time to ask Cranes directly. When nothing applies, it adds nothing. Slash commands and empty messages are skipped.
• Your skills in Claude Code. At the start of each session, it copies your skills from Cranes into Claude Code’s skills folder, named cranes- and the skill’s name, so they load like any other skill and are at most one session out of date. If you edit a copy, it’s never overwritten; Claude offers once to save your edit back to Cranes. Copies of skills you removed from Cranes are removed, unless you edited them. Skills Cranes didn’t write are never touched.
• Everything else in Cranes. The plugin also adds Cranes’s connection, so you can read and change your thinking from a session. See What your AI can do with Cranes.

What still depends on Claude: saving. It offers once, at a natural end like a fix confirmed or a commit, and saves only if you say yes. See What connected AIs keep.

Install the plugin

The plugin is installed from a copy of Cranes’s code for now; it isn’t in a public plugin marketplace yet. From a checkout of the repository, run these in Claude Code:

/plugin marketplace add ./pikespeak/plugins/claude-code
/plugin install cranes@cranes

Then restart Claude Code, or run /reload-plugins.

Sign in

The plugin has two sign-ins, because its automatic parts can’t use Claude Code’s own connection:

1. For your thinking on every prompt and your skills: run /cranes:login. Your browser opens Cranes; choose what the plugin may use and choose Allow. Claude Code says in a line whether you’re signed in. The sign-in renews itself.
2. For everything else: the first time Claude uses Cranes for anything else, like saving or reading a page, Claude Code asks you to connect Cranes. You can also do it from /mcp.

Both appear in Settings › Connected AIs, the first as Cranes for coding agents. That first sign-in is shared on your computer with the Cranes plugins for Cursor and Codex, so you sign in once for all of them. Until you run /cranes:login, the plugin adds nothing to your prompts.

What it sends

With each prompt, the plugin sends your message and a short description of the work: the repository and branch, whether you have uncommitted changes, how the session started, and up to five of your earlier messages in the session that corrected Claude, word for word. It never sends your code, files, or Claude’s replies. The description is used only to find what applies, and is never kept or learned from.

If Cranes is slow or can’t be reached, the plugin adds nothing and your prompt goes ahead as usual.

Only the connection

To add Cranes without the plugin, run this in a terminal, then run /mcp in Claude Code, choose Cranes, and sign in:

claude mcp add --transport http cranes https://usecranes.com/connect/mcp

Claude Code then has everything in What your AI can do with Cranes, but calls Cranes only when it decides to. To make that more reliable, add the line from Have your AI use Cranes on its own to your repository’s CLAUDE.md. See When Cranes comes in.