Claude Code plugin
The can plugin is the shortest path from Claude Code to your boards. It bundles the hosted MCP server — so there is no URL to paste and no package to configure — and adds the /can:pr workflow command on top.
Install
Run two slash commands inside a Claude Code session. First, add the Random Fact Labs plugin marketplace:
/plugin marketplace add randomfactlabs/claude-plugins
Then install Can from it:
/plugin install can@randomfactlabs
Both commands sit behind copy buttons in Can, under Workspace settings → Connections.
Authenticate
Installing the plugin registers the Can MCP server; it does not sign you in. That step is yours, once:
- Run
/mcp. - Pick Can and finish the sign-in in the browser tab that opens.
From then on tokens refresh on their own. You won't see the browser again unless access is revoked or you sign out.
Do this in an interactive session before you rely on Can in a headless one — claude -p and Agent SDK runs have no /mcp panel to complete the sign-in with.
Use
Ask for your boards in plain language — "list my Can boards", "what's in Review on LAUN?" — and Claude reaches them through the same curated tool catalog every hosted-MCP client uses.
The plugin adds one command of its own:
/can:pr CAN-123
That takes a ticket from the board to a review-ready pull request. Claude reads the ticket, assigns it to itself and to you, moves it to an in-progress column, writes the change on a branch, runs your project's checks, opens the PR, and comments the link back on the ticket. The ticket stays in-progress — a pull request is a request for review, not acceptance — and moves to done once it merges.
Hand it several keys at once (/can:pr CAN-123 CAN-124) and it fans out: one subagent and one pull request per ticket, running in parallel.
The workflow itself is served by the MCP server's get_help tool, so it tracks changes in Can without you reinstalling the plugin.
Which connection do I want?
| If you are | Use |
|---|---|
| In Claude Code | This plugin |
| In claude.ai, ChatGPT, Cursor, or another MCP client | The hosted MCP server |
| After every API operation, or a stdio-only setup | The local MCP package |
| Writing your own integration | The REST API |
They all authenticate the same way and land in the same place: the agent acts as you, your workspace roles apply unchanged, and every change it makes is recorded in the ticket's activity history.