cc-connect

Purpose

cc-connect is a bridging tool that connects AI coding agents running on a local machine to the instant messaging platforms users use every day. It lets users remotely control local AI agents through a chat app without having to stay at the terminal, completing tasks such as code review, research, automation, and data analysis.

Main Capabilities

• Universal agent support: Supports more than 10 AI agents, including Claude Code, Codex, Cursor Agent, Kimi CLI, Qoder CLI, Gemini CLI, OpenCode, iFlow CLI, Pi, Devin, Copilot, and any agent that supports the Agent Client Protocol (ACP).
• Multi-platform access: Supports 13 chat platforms, including Feishu, WPS Collaboration, DingTalk, Slack, Telegram, Discord, WeCom, Weibo, LINE, QQ, QQ Bot (official), Matrix, and personal WeChat (Weixin personal ilink). Most platforms do not require a public IP address.
• Multi-agent orchestration: Multiple bots can be bound in a group chat and allowed to communicate with each other, for example asking Claude a question and getting insights from Gemini in the same conversation.
• Full control within chat: Switch models via slash commands (/model), adjust reasoning (/reasoning), change permission mode (/mode), and manage sessions; use /dir <path> to switch directories (/cd <path> is a compatibility alias), and /dir <number> or /dir - to quickly jump through history.
• Persistent memory: Read and write agent instruction files via /memory without touching the terminal.
• Smart scheduling: Set cron tasks in natural language, for example "summarize GitHub trends every morning at 6 o'clock."
• Multimodal support: Send voice messages or screenshots, and cc-connect handles STT/TTS and multimodal forwarding.
• Multi-project architecture: One process supports multiple projects, and each project can have its own combination of agents and platforms.
• Multilingual interface: Native support for English, Simplified Chinese, Traditional Chinese, Japanese, and Spanish.

How to Use

1. Install the AI agent CLI: Install at least one agent to bridge, and make sure its binary is in PATH. For Cursor Agent and OpenCode, follow the official installation pages.
2. Authenticate the agent: Each agent has its own login flow, and it must be run interactively once to store credentials.
3. Install cc-connect: You can use Homebrew (macOS/Linux), download the binary from GitHub Releases, or build from source (requires Go 1.22+).
4. Start and open the Web UI: On first startup, a URL like this will be printed; open it in a browser. If port 9820 is occupied, you can pass --web-port 9821 or set web_port in config.toml. Note that cc-connect web only opens the browser and configuration interface; it does not start the service, so you still need to run cc-connect in another terminal.
5. Configure the platform bot token in the Web UI: Create a project, add at least one platform, paste the bot token from the corresponding platform's developer console, and after saving, cc-connect will hot reload.

You can also install and configure it through an AI agent (recommended), or configure it manually. For manual configuration, you can set admin_from = "alice,bob" in the project to allow specific user IDs to run privileged commands (such as /dir and /shell); admin_from must be placed under [[projects]]. Use /whoami or /status to get the current user ID. Run /dir reset to restore the configured work_dir and clear persistent overrides.

Use cases

• Interact remotely with a local AI coding assistant via phone or tablet.
• Coordinate multiple AI agents working together in a group chat.
• Set up scheduled tasks through natural language to automatically perform operations such as summaries and checks.
• Need to connect to a chat platform without exposing a public IP.
• Develop multiple projects in parallel, with each project using a different combination of agent and platform.

Limitations

• You must first install and authenticate at least one AI agent CLI, otherwise cc-connect will exit and report that the CLI cannot be found.
• WeCom Webhook mode requires a public URL; long-connection/WS-style setups usually do not.
• Voice features need to be enabled in config.toml with [speech] / TTS provider.
• OS user isolation (run_as_user) is currently supported only by Claude Code, and requires conditions such as passwordless sudo for the target user.
• The default attachment sending limit is 50 MiB, adjustable via max_attachment_size_mb or an environment variable.
• Sessions automatically rotate after a long period of inactivity (default 30 minutes), and this can be disabled with reset_on_idle_mins = 0.
• Some platforms (such as QQ NapCat/OneBot) are unofficial self-hosted bridges, and behavior depends on NapCat/network settings.
• Details such as specific charges, pricing, and the full platform support matrix are not yet provided.