CLAUDE.md files and your skills work in Gloo Code as they are. You only convert your MCP servers, which takes a few minutes.
What carries over automatically
Type
/skills inside Gloo Code to see the skills it found. Each skill is also available as /<skill-name>, and the agent can use a skill on its own when the task matches its description.
AGENTS.md and CLAUDE.md load either way. Project config, skills, commands, and MCP servers from the folder apply after you trust it. If the folder has its own config (a gloocode.json, or a .gloocode, .claude, or .agents folder), Gloo Code asks once, and you can use /trust whenever you like.Do once
1
Install and sign in
Follow Installation & Setup. Sign-in happens in your browser, and your organization needs a Gloo Code subscription with a seat for you.
2
Open your project and start a session
Run
gloocode in the folder you normally use Claude Code in, and trust the folder when asked.3
Optional: tidy your instructions with /init
If you’d like one tool-neutral file, run
/init. Gloo Code reads your existing CLAUDE.md and other rule files and drafts an AGENTS.md you can review. Keeping CLAUDE.md as it is works fine too.4
If CLAUDE.md pulls in other files with @
List those files under
instructions in a gloocode.json at your project root, and Gloo Code loads them as instructions.gloocode.json
5
Convert your MCP servers
Copy each entry from
.mcp.json into the mcp section of gloocode.json. The recipe is on Move your MCP servers.6
Bring over custom slash commands
Copy your command files into Optional frontmatter keys are
.gloocode/commands/. Each Markdown file becomes a command named after the file, and $ARGUMENTS, $1, and $2 placeholders in the body work as you’d expect.description, agent, model, and subtask.Day to day
Handy to know: Ctrl+X is Gloo Code’s leader key, used for chords such as Ctrl+X then N for a new session, L for sessions, A for agents, M for models, U to undo, and R to redo. Shift+Enter adds a new line, and keybindings can be changed in
tui.json.
How work flows
You describe a task, and the build agent coordinates it. It hands work to subagents such ascode, explore, review, and qa, which can run in parallel, and shows each as a card in the session so you can follow along. /review and /qa run those subagents on demand.

