Connect your agent

Claude Code, Cursor and other MCP clients; identity and permissions.

Corpole is useful the moment your coding agent can read it. The agent connects over the Model Context Protocol (MCP) to one project, reads what binds for the work in front of it, and proposes what it finds. This section covers every way to make that connection and what each one allows.

Claude Code gets the most from Corpole: with the plugin it loads the binding records at session start, shows Corpole's notices in the terminal, stops before edits to ungoverned code and offers the /corpole:* commands. Cursor, VS Code and other MCP clients connect with a URL and, where they do not support sign-in, a personal token. claude.ai, Claude Desktop and ChatGPT connect as a custom connector and are well suited to planning, interviews and reviews.

Identity matters as much as connectivity. Every connection acts for exactly one person and never has more permissions than that person; everything it does is recorded under their name. The articles on tokens and on read-only versus propose explain who can do what, and why an agent can propose but never decide.

Finally, the repository has to agree that Corpole comes first. /corpole:init writes the small set of files that make Claude Code rank Authority records above the documents in your repository, and adds the pull-request check if you want it.

Articles in Connect your agent