Use Claude Code in Mezair
Mezair Desktop can run Claude Code against your local repositories using your
installed claude CLI and its existing sign-in. Choose Claude Code for a chat,
make it your primary provider, or assign it to a local workflow step.
Connect your CLI
Install Claude Code using Anthropic's setup guide. If you already use it in a terminal on this computer, keep that installation.
Confirm it is available and sign in:
claude --version claude auth login claude auth statusIn Mezair, open Settings → Providers & models → AI connections and enable Claude Code. Make it primary if you want it for new chats and app-level AI tasks.
Choose Check again to refresh sign-in status and available models. You can also start Claude's browser sign-in from this panel.
Mezair uses the official Anthropic Agent SDK with your installed executable. Authentication stays with Claude Code; Mezair does not copy its OAuth files into its database or send them to another machine. You do not need to enter an Anthropic API key into Mezair for this login path.
Choose a model
The chat model picker, primary model field and local workflow editors offer the models reported by your CLI. You can also enter a custom model ID.
| Choice | Behavior |
|---|---|
Use CLI default (auto) |
Keeps the model from your Claude Code configuration. A CLI configured for Haiku keeps using that preference. |
| A listed model | Uses that model for the chat or workflow step. |
| A custom model ID | Passes the selected ID to Claude Code. Availability depends on your account. |
Choose any model your Claude account supports. Model access and usage limits follow your Claude plan. Model discovery and sign-in checks do not send model prompts; Test connection sends a small request using the selected model.
Claude turns do not use Codex's reasoning-effort or priority-tier controls.
Work in a local chat
Claude Code chats support streamed replies, tool calls and results, successful file-change evidence, approvals and structured questions. Persona instructions and repository context accompany the session. Follow-up messages use Queue while a turn is running; you can interrupt the turn or stop the session.
Reopening a Claude session on the same computer uses its native session ID. Branching starts a new Claude thread with conversation context.
| Permission choice | Claude behavior |
|---|---|
| Agent | Uses Claude's default permission checks. |
| Accept edits | Applies file edits without asking and keeps Claude's other permission checks. |
| Full access | Bypasses Claude's permission checks. Choose this only when you intend that access. |
| Plan | Uses Claude's planning mode. This is not a filesystem sandbox. |
Workflows and automations
Enable Claude Code before assigning it to a local workflow step or specialist. Each step stores its own provider and model. Persona automations use the primary provider and its selected model.
These background runs use claude --print. They record the final text without
the interactive chat's tool, diff or reasoning events. Review their output and
repository changes before keeping the work.
Current limits
- Claude Code execution and login stay on the local Desktop. Browser execution, cloud execution and remote credential delivery do not support this provider.
- Strict read-only sessions, read-only side questions and temporary chats are unavailable for Claude Code. Plan mode does not provide the same isolation.
- Native goal controls and mid-turn Guide are unavailable; use Queue for follow-ups.
- Software factory workers currently require Codex.
- Provider availability still depends on the installed CLI, its login and your Claude account's model access and remaining usage.
Ask your coding agent to help
Help me connect the Claude Code CLI on this computer to Mezair Desktop, following
https://mezair.dev/docs/desktop/claude-code. Check whether claude is installed and
signed in, then explain how to enable it under Settings → Providers & models.
Keep my current CLI model as the default unless I choose another model. Show me
how to select available models in chat and local workflows. Do not send a test
prompt or change my Claude configuration unless I ask. The expected result is
a connected local provider with my model preference preserved.