Skip to main content
Claude Code is Anthropic’s official terminal-native coding agent powered by Claude.

Install Claude Code

You can skip this section when using the setup wizard below. If it cannot find the claude command, it offers to run the official Claude Code installer. Refer to the Claude Code documentation for installation.

Configure MiniMax API

Before configuring, clear these environment variables. They take priority over the settings file and override the MiniMax configuration:
If ~/.bashrc or ~/.zshrc exports them, remove those lines too, or new shells will load them again.
Run:
Select Claude Code. If Claude Code isn’t installed, select it again when prompted to choose which tools to install. The wizard runs the official installer, checks claude --version, updates ~/.claude/settings.json, and adds the MiniMax model list.See MiniMax CLI for command options and backup behavior.
To add web search, see Web Search MCP.

Verify the configuration

After starting claude, run the following slash commands inside the TUI to confirm Claude Code is using MiniMax instead of the default Claude provider:
  • /status should show the ANTHROPIC_BASE_URL pointing to api.minimax.io/anthropic.
  • /model should show the active model as MiniMax-M3.1-Flash.
Thinking is always enabled for MiniMax-M3.1-Flash and cannot be disabled. Reasoning effort controls thinking depth. Choose from low, medium, high, xhigh, or max. Higher reasoning effort can increase response time and usage.
  • Default level: set by CLAUDE_CODE_EFFORT_LEVEL in env of ~/.claude/settings.json. The Option 2 and Option 3 examples set it to max. When it is not set, Claude Code uses high.
  • Switch during a session: run /effort to pick a level, or run /effort <level> directly, for example /effort low. You can also start with claude --effort <level>, which applies to that session only. While CLAUDE_CODE_EFFORT_LEVEL is set, /effort in a session has no effect; remove the variable first if you want to switch in a session.
  • Some Claude Code versions send xhigh as high; use max if you need deeper thinking.
  • Settings that turn thinking off, such as MAX_THINKING_TOKENS=0 or "alwaysThinkingEnabled": false, have no effect on MiniMax-M3.1-Flash; the model still thinks.