← Discover MCPs and Agents
F
MCPAI & MLGitHub

Fuxi

FuXi is a fast, self-contained AI coding agent that lives in your terminal — edit code, run commands, and drive tools, with cost-aware routing across LLM providers.

Links

README

From the repo.

FuXi

English | 简体中文

GitHub stars Release Last commit License

FuXi is a fast, self-contained terminal AI coding agent. It reads code, edits files, runs commands, and drives tools from a rich TUI, with cost-aware routing across LLM providers and automatic failover. Built in Go, it ships as one static binary with no runtime dependencies.

Terminal-first · Provider-agnostic · Bring your own key · MCP client · Self-updating

Homepage: https://www.fuxicode.com

FuXi's roughly 2 million lines of code were written entirely by FuXi itself.

FuXi in action

Install

macOS / Linux

curl -fsSL https://downloads.fuxicode.com/bootstrap.sh | bash

Windows (PowerShell)

irm https://downloads.fuxicode.com/bootstrap.ps1 | iex

Windows (CMD)

curl -fsSL https://downloads.fuxicode.com/install.cmd -o "%TEMP%\fuxi-install.cmd" && "%TEMP%\fuxi-install.cmd"

Then verify and launch:

fuxi --version
fuxi doctor    # environment sanity checks
fuxi           # start a session

Install and upgrade are the same command. Full options, uninstall steps, and troubleshooting: setup docs.

Get started

A FuXi account is required. Run fuxi, then:

  1. Register and sign in — fuxi login (or fuxi setup-token for headless/CI). Registration processes only minimal account data.
  2. Connect a model (optional) — sign-in gives access to FuXi-managed models; to use your own provider instead, set an API key via environment variable or ~/.fuxi/config.yaml (fuxi init generates a template). Or run fuxi wizard.

Then type a prompt and press Enter, for example:

Fix the failing tests in this repository.

Useful commands: /model switch models · /help browse all commands · /config open settings · /privacy-settings privacy controls · /exit quit.

Full walkthrough: usage guide.

Permissions

Sensitive tool calls ask for approval unless you change the permission mode at launch.

--dangerously-skip-permissions starts in bypassPermissions, so tool calls run without a prompt. Extreme commands such as rm -rf / are still blocked. Use it only in a container or VM; on Linux and macOS, root and sudo are rejected.

--allow-dangerously-skip-permissions keeps the normal mode and adds bypassPermissions to the Shift+Tab cycle.

--permission-mode bypassPermissions behaves like --dangerously-skip-permissions. The same flag also accepts default and plan.

Data, privacy, and retention

FuXi runs on your machine and keeps your work local.

  • By default, your code and conversations are never collected, stored, or retained by FuXi. With your own key they go straight to the provider you choose; with FuXi-managed models they are transmitted only to serve that request. Conversation content is sent to FuXi only if you explicitly enable send_conversations (off by default).
  • Credentials stay in local ~/.fuxi/ and are never uploaded.
  • Sessions, checkpoints, memory, and audit logs live on your device and can be deleted at any time by removing the config directory.
  • FuXi processes limited account and technical information to operate the service; the available analytics are disclosed and can be turned off. See Privacy Controls.

Full details: Privacy Policy · Security Whitepaper · Terms of Service.

Documentation

GuideWhat it covers
Usage guideFirst session, permissions, sessions & memory, tools & MCP, CLI reference, troubleshooting
ArchitectureExecution base, persistence, routing, multi-window coordination, extensibility
Keyboard shortcutsTerminal-UI key reference
Environment variablesFull environment-variable reference
FAQCommon questions
Benchmark reportReproducible evaluation methodology and results
Security & privacyPrivacy Policy, Security Whitepaper, Terms of Service, Usage Policy, compliance and governance
ChangelogRelease history
SupportWhere to get help

Usage policy

You must use FuXi in accordance with the laws applicable to you. FuXi is aimed at developers and professional users and is not directed at children under 14; users aged 14–17 require a parent or guardian.

Disallowed uses and full requirements: Usage Policy. If you are unsure whether a use is lawful, take advice before proceeding.

Contributing

The product source is proprietary; this repository hosts the documentation, installers, and issue tracker, which are open to contributions. See CONTRIBUTING.md.

License

Proprietary. Copyright © 2026 FUXI. All rights reserved.

Collected info

  • ★ 3,366 stars
  • ⎇ 227 forks
  • Language: Python
  • Source updated: 9/25/2026

Config for your environment

Replace {MCP_ENDPOINT_URL} with this MCP’s endpoint URL (from its repo or docs above). No API key — you connect directly.

Tool

OS

Config file: ~/.cursor/mcp.json

{
  "mcpServers": {
    "mcp-server": {
      "url": "{MCP_ENDPOINT_URL}"
    }
  }
}

Paste into mcpServers in the config file. Restart Cursor after saving.

If this MCP is also published on mcpchannel.ai, you can subscribe from Browse and use the gateway config there instead.