Who cli is for#
Developers building AI agents that operate Lark
The 26 AI Agent Skills give an LLM a structured, minimal surface to send messages, create calendar events, read documents, update Base records, and summarize meetings without requiring the agent to parse raw Lark API responses. Authentication and scope selection are handled at setup, so the agent only runs approved operations.
Skip if:
Your AI agent workflow runs entirely outside of Lark or Feishu. lark-cli is the wrong tool if your agent operates on other platforms like Slack, Notion, or Google Workspace.
DevOps engineers automating Lark-based workflows
lark-cli's consistent JSON output, pagination flags, and dry-run mode make it practical for shell scripts and cron jobs. You can paginate through all messages in a channel, export Base records to ndjson for downstream processing, or send status notifications to a Messenger chat from a deployment pipeline.
Skip if:
Your team uses a visual workflow builder and does not want to maintain terminal scripts. Zapier or Make may be faster to configure if shell scripting is not part of your stack.
Enterprise teams embedding Lark automation in internal platforms
The extension/ packages let enterprise IT and ISVs wrap lark-cli in their own binary, inject credentials from a Vault or config center, restrict the available command surface, and route all API calls through a centralized audit log without modifying the CLI source.
Skip if:
Your use case is a single developer running commands interactively. The enterprise extension path adds setup complexity that personal use does not need.
The problem it solves#
Automating tasks in Lark or Feishu without a dedicated CLI means reading through dense API documentation, managing OAuth token flows by hand, parsing inconsistent response shapes, and writing custom wrappers for every operation your scripts need. Most teams either build and maintain fragile shell scripts against raw REST endpoints, or pay for a visual automation platform that charges per-task and limits what the Lark API actually exposes.
The problem gets sharper for AI Agent workflows. LLMs need predictable, structured outputs and minimal surface area to operate a tool reliably. Raw Lark API responses require parsing and preprocessing before an agent can act on them reliably, and there is no standard contract for errors, pagination, or scope requirements. Building a Lark-integrated agent from scratch means solving all of these before you can write a single useful agent instruction.
How it solves it#
200+ commands across 18 business domains
Covers Calendar, Messenger, Docs, Sheets, Base, Mail, Meetings, Tasks, Wiki, Contacts, Approval, OKR, Attendance, Drive, Markdown, Slides, and more. Commands are generated from Lark OAPI metadata and curated through quality gates, each mapped 1:1 to platform endpoints.
26 AI Agent Skills for structured automation
Each Skill is a curated set of commands for a specific Lark domain, designed for use with LLMs and AI agents. Skills ship structured JSON output, enforce secure defaults, and include scope metadata so agents can request exactly the permissions they need without over-provisioning.
Three-layer command architecture
Shortcut commands (prefixed with +) provide human and agent-friendly access with smart defaults and table output. API commands map 1:1 to Lark endpoints. Raw API mode covers 2,500+ Lark Open Platform endpoints for cases that fall outside the curated command set.
Multiple output formats for scripting and pipelines
Supports output formats including json (default), pretty, table, ndjson, and csv. JSON output uses a consistent success/error envelope where success goes to stdout with exit code 0 and errors go to stderr with a typed error object containing the upstream API code, subtype, and a human-readable hint.
Input injection protection and OS keychain credential storage
Credentials are stored in the OS-native keychain rather than plain config files. The CLI sanitizes terminal output and applies input injection protection, which reduces risk when AI agents invoke it with untrusted data sourced from messages or documents.
Strengths and trade-offs#
Strengths
- MIT license with no per-task pricinglark-cli is MIT licensed, so there are no licensing fees or per-execution charges. Unlike paid automation platforms like Zapier or Make that bill per task or per month for Lark integrations, running lark-cli costs nothing beyond the machine running it.
- Maintained by the official larksuite teamThe CLI is built and maintained by Lark's own engineering team and keeps pace with the Lark Open Platform API. Commands are generated automatically from official OAPI metadata. With 17,000+ GitHub stars and last pushed September 2026, it is actively developed, not a community fork.
- Agent-native design tested with real AI agentsEvery command is tested against real AI agents with concise parameters, smart defaults, and structured outputs tuned for high LLM call success rates. The 26 Skills expose exactly the operations agents need without giving them access to the full API surface they do not require.
- Full 2,500+ endpoint coverage via raw API modeBeyond the curated 200+ commands, raw API mode lets you call any Lark Open Platform endpoint directly. This means no waiting for a new command when a new Lark API is released; the raw access layer covers it immediately with the same authentication and output contract.
Trade-offs
- -Lark and Feishu platform onlylark-cli works exclusively with Lark and Feishu APIs. If your team uses Slack, Microsoft Teams, or a different collaboration platform, this tool provides no value. There is no cross-platform automation layer or support for other messaging or collaboration services.
- -Requires app credentials and OAuth setup before first useBefore any command works, you need to create a Lark app in the Open Platform console, configure its credentials, and complete an OAuth login. The interactive setup takes a few minutes for personal use, but teams with centralized credential management need to review the enterprise extension path before deploying in production.
- -693 open issues at time of reviewThe GitHub repository carries 693 open issues. For a tool created in March 2026, that number suggests the issue tracker is active but some commands or edge cases may have unresolved bugs. Check the issue tracker for your specific domain before depending on lark-cli for critical automations.
cli vs alternatives#
lark-cli vs Zapier for Lark Automation
Both tools can automate Lark operations, but they serve different workflows. lark-cli is a code-first CLI for developers and DevOps engineers; Zapier is a visual automation platform with a Lark connector that non-technical users configure in a browser without writing code.
| Feature | lark-cli | Zapier |
|---|---|---|
| License | MIT | Proprietary |
| Pricing | Free | $19.99/month+ for multi-step automations |
| Lark API coverage | 200+ commands, 2,500+ raw endpoints | Limited to connector actions |
| AI Agent support | 26 Agent Skills, purpose-built | Via Copilot AI features |
| Setup | npm install + OAuth | Browser-based, no code required |
| Output format | Structured JSON with typed error contract | Webhook payloads or mapped fields |
lark-cli is the better choice when you need full API coverage, scripting or CI integration, AI agent workflows, or when per-task billing would make automation expensive at scale. Developers get direct access to every Lark endpoint the Open Platform exposes, with consistent output contracts suited for pipeline integration.
Zapier remains the better fit when non-technical team members need to build automations without writing code, or when you need to connect Lark to third-party services. Zapier's no-code interface has a shorter time-to-first-automation for simple trigger-then-action workflows.
lark-cli vs Make for Lark Integration
Make (formerly Integromat) is a visual automation platform with more advanced scenario-building than Zapier, but the same fundamental tradeoff applies compared to lark-cli: no code required, per-operation pricing, limited Lark API coverage, and no native AI Agent workflow support.
lark-cli outperforms Make when the use case is a developer writing scripts, building agents, or needing access to Lark APIs that Make's connector does not expose. Make is the stronger choice for teams that need complex multi-step workflows built and maintained by non-engineers who prefer a visual interface.
Quick start#
Install lark-cli from npm, then configure app credentials and authenticate.
```bash
npx @larksuite/cli@latest install
lark-cli config init
lark-cli auth login --recommend
```What it's built on#
- Languages
- GoJavaScriptPython
FAQ#
Is lark-cli free to use?
Yes. lark-cli is MIT licensed and free to use for any purpose, including commercial use. There are no licensing fees, per-task charges, or usage limits imposed by the CLI itself. You need a Lark or Feishu developer app to authenticate, but the CLI adds no cost on top of that.
Does lark-cli work with Feishu as well as Lark?
Yes. Lark and Feishu are the same product sold under different names in different regions. lark-cli supports both and the same credentials, commands, and Skills work on either platform. The GitHub repo and README use both names interchangeably.
How do I authenticate lark-cli for the first time?
Run lark-cli config init to walk through creating a Lark app and configuring credentials interactively. Then run lark-cli auth login to authenticate. Credentials are stored in your OS-native keychain. Enterprise deployments with centralized credentials should review the extension/ packages for a non-interactive path.
Can AI agents use lark-cli without human intervention?
Yes. The CLI ships with 26 AI Agent Skills and includes a non-blocking auth login mode (--no-wait) so agents can initiate OAuth without blocking on user input. The Lark team tests every command against real agents, and the structured JSON output contract is designed for LLM parsing. The README includes explicit Quick Start steps for AI agents.
What is the difference between Shortcuts, API commands, and raw API calls in lark-cli?
Shortcuts (prefixed with +) are human and agent-friendly commands with smart defaults, table output, and dry-run mode for safe preview. API commands map 1:1 to Lark Open Platform endpoints and are auto-generated from official metadata. Raw API mode lets you call any of the 2,500+ Lark endpoints directly using its URL and HTTP method.
Similar open-source tools#
FckSignups
Open-source tools that work instantly, no signup required
FreeFlow
Free, open source Mac dictation with AI cleanup and voice macros
rakazo
AI teammates you own: your keys, your model, your machine.
Multica
Assign GitHub issues to AI coding agents. They run on your machine.
LibreChat
One self-hosted interface for every AI model you use
opendisplay
Turn any iPhone, iPad, or spare Mac into a free second Mac monitor