wechat-claude-code: talking to Claude Code over WeChat

What the bridge actually does

The premise is simple. After you bind WeChat by scanning a QR code, a new contact appears in your WeChat. You send it a message, the message is forwarded to the Claude Code running on your computer, and the reply is pushed back to WeChat in real time. The README lists support for sending and receiving text, images, voice and files. The design goal stated in the project title is to chat with Claude Code in WeChat just like texting a friend. The project is MIT licensed, the primary language is TypeScript, the default branch is main, the last push was on 2026-06-27, and there are no GitHub releases. It is packaged as a Claude Code Skill, with a SKILL.md at the repository root and a skills.sh listing badge, which is why the installation instructions are framed around the skills CLI rather than a package manager.

The architecture: four hops, all local except the WeChat side

The README draws the data flow as a single line: the phone on WeChat, the ilink Bot API, the Node.js daemon, and the Claude Code CLI on the local machine. The daemon listens for WeChat messages by long polling, forwards them to the local claude CLI, and streams the reply back to WeChat. The README emphasises that the whole thing runs on your own computer. The distinction that matters for an adopter is where the trust boundary sits. Your messages travel through a platform API, but the code that talks to Claude Code, the working directory it operates in, and the credentials it uses never leave the machine. The daemon is the component that holds all of that, and it is the component you have to keep running.

Installing the skill, two ways

The recommended route is the skills CLI, and the first trigger in a conversation will clone the project source and install its dependencies automatically.

npx skills add Wechat-ggGitHub/wechat-claude-code

The manual route clones the skill into the skills directory of your home folder and installs the dependencies there.

git clone https://github.com/Wechat-ggGitHub/wechat-claude-code.git ~/.claude/skills/wechat-claude-code
cd ~/.claude/skills/wechat-claude-code && npm install

The package is an ES module TypeScript project, and the build is a plain tsc invocation, with a postinstall hook that runs the build for you. The runtime dependencies are small, qrcode and qrcode-terminal, both there to draw the pairing code.

Binding the account and running the daemon

The first step after installation is setup, which is what produces the QR code you scan in WeChat.

cd ~/.claude/skills/wechat-claude-code
npm run setup

A QR code appears and you scan it. Then the daemon is started, and on macOS the README says it registers a launchd agent automatically, so it starts at boot and restarts after a crash.

npm run daemon -- start

From there you open WeChat and send a message to the new contact. The daemon wrapper is a bash script, scripts/daemon.sh, invoked through the daemon script in package.json, and it accepts the subcommands status, stop, restart and logs, with restart being the one to use after you pull new code.

The command set you can send from the chat window

The README documents a set of slash commands that are sent as ordinary chat messages, which is what makes the bridge feel like a chat rather than a control panel. There is /help to show the help text, /clear to clear the current session and start a new conversation, /stop to stop the running task, /model followed by a name to switch Claude model, /prompt followed by content to set the system prompt, /cwd followed by a path to change the working directory, /skills to list the installed Skills, /status for the current session state, /history with an optional count for recent conversation records, /compact to compact the context and start a new CLI session, /reset to reset everything including the working directory setting, /undo with an optional count to undo recent turns, and /<skill> with parameters to trigger any installed Skill. The combination of /cwd and /prompt is the pair that gives the bridge its power, since it lets you point a task at a particular project and give it instructions in a particular language.

What the bridge filters out, and what it shows you

Two design choices are worth understanding before you rely on it. The first is message filtering. The README lists a feature it calls no screen spam, meaning only core information is pushed, namely progress, results and key decisions, while tool calls and intermediate process are filtered out automatically, and it frames this as keeping the reading experience clean. The second is presence. There is a feature that shows a typing indicator in the chat, so you can tell that Claude is working on the task while you wait. A third is the timeout reassurance, described in the README as a task that goes more than five minutes without a response sending a message to say it is still working, so you are not looking at a blank chat box. The file handling goes both ways as well, so you can send an image, a Word document or a PDF for analysis and receive files Claude generates directly in the chat.

Where the data lives, and the known rough edges

The data directory is documented in full, and the layout tells you what the daemon keeps. Everything is stored under ~/.wechat-claude-code/: an accounts/ directory holding the WeChat account credentials, config.json for global configuration, sessions/ for session data, and logs/ for run logs, which rotate daily and are kept for 30 days. The rough edges are also documented. The README's future plans section opens with message queue optimisation, noting that when several commands are sent in a row the replies can get mixed up, and that a better queue strategy is being researched. The second planned item is not interrupting the computer when it sleeps, using the macOS caffeinate command so the machine stays awake and the bridge still answers with the lid closed. The third is continuing a session, so that a conversation started at the desk can be picked up on the phone with the same workspace and context. Those are the three things to expect to be imperfect today, and the first is the one you will hit first.

Requirements, and the licence and maintenance picture

The prerequisites are four lines: Node.js 18 or newer, macOS or Linux, a personal WeChat account, and Claude Code already installed and authenticated. There is also a note that Claude Code supports third party API providers such as OpenRouter and AWS Bedrock, in which case you set ANTHROPIC_BASE_URL and ANTHROPIC_API_KEY. Windows is not in the supported list, and the launchd registration is macOS specific, so the practical reading is that this is a macOS and Linux tool today even though the daemon script is plain bash. The licence is MIT, with the LICENSE file at the repository root, and the project ships both a Chinese and an English README, README_en.md, plus a docs/ directory and a scripts/ folder. The repository is not archived, and with the last push on 2026-06-27 and no releases, you are tracking the default branch rather than a tagged version.

Editorial conclusion

Adopt wechat-claude-code if you drive Claude Code from a desktop and want to hand it a task, or read a result, from your phone over WeChat, and if you are willing to bind a personal WeChat account to a local daemon. Do not adopt it for shared or work accounts, since the binding is to a personal account and the README states the data stays on your machine, and do not expect a Windows path, because the prerequisites list macOS or Linux only. Before you install, read the daemon subcommands so you know how to stop it, and check the message queue note, because the README says consecutive commands can interleave their replies.

Frequently asked questions

Official sources

1. Issues
2. License: MIT
3. README
4. Wechat-ggGitHub/wechat-claude-code on GitHub