Feishu (Lark) recently open-sourced its official CLI, lark-cli, so I hooked it up to Multica. Now the agents in Multica can handle Feishu messages, docs and calendar events directly. Here is how I set it up.
What Is lark-cli
lark-cli is the official command-line tool maintained by the Feishu/Lark team (MIT licensed), built so that “both humans and AI agents can operate Feishu from the terminal”. It covers messaging, docs, Base, Sheets, calendar, mail, tasks, meetings, approvals, OKRs and more, with 200+ commands and twenty-plus ready-made AI Agent Skills (such as lark-im, lark-doc, lark-calendar and lark-task).
That matters for Multica: Multica itself needs no Feishu-specific integration. As long as lark-cli and its Skills are installed on the machine that runs the agents, agents like Claude Code and Codex will call lark-cli on demand while working on tasks.
How the Integration Works
Multica agents are CLIs (Claude Code, Codex, etc.) launched by the local daemon on the runtime machine. So “connecting Multica to Feishu” comes down to two things:
- Install lark-cli and its Skills on the machine running the Multica daemon;
- Complete the Feishu app setup and login as the same OS user.
After that, assign an issue to an agent in Multica, e.g. “Summarize today’s discussion in the Feishu group and write it up as a doc”, and the agent will call lark-cli by itself.
Installation
Prerequisite: Node.js is installed (npm/npx available).
npx @larksuite/cli@latest install
This install command is an interactive setup wizard that:
- Installs
@larksuite/cliglobally (giving you thelark-clicommand); - Installs the AI Skills (equivalent to
npx skills add larksuite/cli -y -g, into the user-level Skills directory where agents like Claude Code can discover them); - Configures a Feishu app (equivalent to
lark-cli config init --new; just follow the prompt to create the app in your browser); - Asks whether to authorize the AI to access your Feishu data on your behalf.
If step 3 or 4 fails or is skipped, simply run the corresponding command manually later.
Login and Authorization
lark-cli auth login --domain all
--domain all requests permissions for every business domain (messages, docs, calendar, tasks, mail, etc.) at once, so you don’t have to keep re-authorizing whenever an agent needs a new feature. The command prints an authorization link; open it in a browser and confirm. This works fine on a headless server too.
If you’d rather grant fewer permissions, narrow the scope:
# Authorize calendar and tasks only
lark-cli auth login --domain calendar,task
Check the status after authorizing:
lark-cli auth status
Verification
First run a command manually in the terminal to confirm Feishu access works:
lark-cli calendar +agenda
If you see today’s agenda, the pipeline is working. Then go back to Multica and create an issue for an agent, for example:
- Check my Feishu calendar for today and turn it into a to-do list;
- Write the conclusion of this issue into a Feishu doc;
- Send a progress update to a Feishu group.
The agent will load the lark-* Skills automatically and use lark-cli to get it done.
Caveats
- Install on the daemon’s machine, under the same user. If your daemon runs on a cloud server, install and log in on that server; installing it on your laptop won’t help.
- Permissions equal identity. Once authorized, the agent acts as you in Feishu, and the official docs warn about risks like model hallucination and prompt injection.
--domain allis convenient but better suited to personal accounts; for teams or sensitive scenarios, grant only what you need. - Keep the bot out of group chats. The official recommendation is to use the connected Feishu bot as a private assistant, not to add it to group chats or let others interact with it, to avoid permission abuse.
- Upgrading. Running
npx @larksuite/cli@latest installagain checks the version and upgrades; already-installed Skills and configured apps are skipped.
Final Thoughts
lark-cli is very agent-friendly, and connecting Multica to Feishu costs almost nothing. Letting agents take care of Feishu messages, docs and calendar events means a lot less context switching.
