engineering / agent
MiniMax Code
An open-source terminal coding agent with an interactive interface, a headless mode and Agent Client Protocol support, for MiniMax or your own model.
At a glance
- Purpose
- An open-source terminal coding agent with an interactive interface, a headless mode and Agent Client Protocol support, for MiniMax or your own model.
- Runs in
- Local agent with shell
- Operating systems
- macOS, Linux, Windows
- Needs
- Runtime: Node.js 22.19+ (22.x), 24.2+ (24.x), 25 or 26; Needs internet: Yes; Needs accounts or API keys beyond your agent: Not stated
- Files and data
- Reads files: Yes; Writes files: Yes; Sends data to other services: Yes
- Tool type
- Agent
- Delivery
- From GitHub
- Version
- Latest on the main branch
- Access
- Code on GitHub
- Compatibility
- According to the README: macOS, Linux and WSL through the shell installer, and Windows through PowerShell; Node.js 22.19+ (22.x), 24.2+ (24.x), 25 and 26; MiniMax accounts for mainland China or the Global region; custom providers using the openai-completions, openai-responses or anthropic-messages API formats; and editors and clients that support Agent Client Protocol. The README says Alpine and musl Linux are not supported by the one-command installer.
- External costs
- According to the README, you need a MiniMax account with available credits (the Token Plan) or your own compatible model API, and model calls may incur charges. The README says search, media tools and managed connectors are subject to account access and service credits. The README gives no prices.
- Permissions
- The README gives remote installer scripts executed through bash or PowerShell and a global npm installation that permits package and SQLite install scripts. The README describes MiniMax browser sign-in or the user's own API key and provider configuration in the active profile's config.yaml. The README says script installs default to ~/.minimax-code or the Windows user-profile equivalent. The README says mcode init generates or updates AGENTS.md. The README says lightweight mode omits tool schemas. The README's uninstall instructions delete the installation directory and warn that it can overlap user data. The README says the agent can inspect diffs, and the TUI offers permission-mode and Plan Mode controls. The README says script installs persistently update your shell PATH or Windows user Path unless MCODE_NO_MODIFY_PATH is set. The README says the agent can read files and run shell commands and tests under permission and sandbox controls.
- Usage terms
- MIT License, as stated in the repository. See its LICENSE file for the exact terms. The README says first-party code defaults to MIT, existing file-level and package-level licences remain in place, and the LICENSE-STATUS.md and THIRD_PARTY_NOTICES.md files cover dependencies, assets and mcode-tools.
Code on GitHub
MiniMax-AI/minimax-code
MITAn open-source coding agent for your terminal, powered by MiniMax.
Topics: coding-agents harness minimax
Public GitHub data fetched · 0 days old
View on GitHub ↗What it does
The README describes MiniMax Code as a terminal coding agent using a MiniMax account or the user's own model. The README gives mcode as its command and lists interactive TUI, headless and Agent Client Protocol entry points.The README says the agent reads files, inspects diffs and runs shell commands and tests under permissions and sandboxing. The README lists session resume, planning, subagents, plugins and built-in skills, and says search, media tools, MCP and managed connectors depend on account access and service credits.
The README gives a pocket-pet browser app example and says the published source covers the terminal TUI, headless CLI and ACP, excludes desktop-app source, and targets the 0.4.12 source preview.
Who it is for
Developers who want a coding agent in the terminal that can use a MiniMax account or their own OpenAI- or Anthropic-compatible model API.Example tasks
Open your project, start mcode and ask it to find a failing test, fix the implementation and run the relevant tests.Run mcode init . to generate or update project guidance in AGENTS.md.
Add your own model provider using the openai-completions, openai-responses or anthropic-messages API format.
Resume the latest session in the current workspace with mcode --continue.
Build the pocket pet example, then change its requirements with a follow-up prompt.