Feishu / Lark MCP

Overview

lark_tools (from veadk.tools.builtin_tools.lark import lark_tools) lets you integrate an AI agent with Feishu / Lark open capabilities for cloud-document processing, chat management, calendar scheduling, and more. It launches a local MCP Server via npx @larksuiteoapi/lark-mcp, so npx and lark-mcp must be installed.

This tool exposes Feishu capabilities as tools the agent can call. If instead you want to route inbound Feishu bot messages into a VeADK Runner, see the Feishu Channel extension.

Environment & prerequisites

To create an app, see Lark app creation.

Configure the API key for the agent’s reasoning model.
Configure the Lark application App ID, App Secret, and user access token.

Environment variables:

MODEL_AGENT_API_KEY: API key for the agent’s reasoning model
TOOL_LARK_ENDPOINT: Lark service Application ID
TOOL_LARK_API_KEY: Lark application App Secret (the variable name remains API_KEY)
TOOL_LARK_TOKEN: Authorized, unexpired user access token

Install Node.js and npm, then run npm install -g @larksuiteoapi/lark-mcp. Both npx --version and lark-mcp -V must work: importing this tool checks both commands. Enable the required application API scopes and authorize the user. The App Secret and user access token are not interchangeable.

Usage

examples/tools/lark/agent.py

import asyncio

from veadk import Agent, Runner
from veadk.memory.short_term_memory import ShortTermMemory
from veadk.tools.builtin_tools.lark import lark_tools

agent = Agent(
    name="lark_agent",
    model_name="doubao-seed-2-1-pro-260628",
    description="A Feishu bot.",
    instruction="Read Lark data as requested. Do not send messages or modify data without an explicit request.",
    tools=[lark_tools],
)

runner = Runner(agent=agent, short_term_memory=ShortTermMemory())


async def main():
    response = await runner.run(
        "List the calendars I can access without creating or changing events"
    )
    print(response)


if __name__ == "__main__":
    asyncio.run(main())

Verify tool discovery

The prebuilt instance reads credentials at import time and configures a local MCP connection. List the available tools first; names and input schemas depend on the installed Lark MCP version. This instance has no cloud-region switch. CLOUD_PROVIDER does not configure platform differences between Feishu and Lark; use custom MCP for custom endpoints or startup arguments.

import asyncio

from veadk.tools.builtin_tools.lark import lark_tools


async def main():
    try:
        tools = await lark_tools.get_tools()
        print([tool.name for tool in tools])
    finally:
        await lark_tools.close()


if __name__ == "__main__":
    asyncio.run(main())

Overview

Get started

Core components

Observability

Extensions

Frontend

Security

Deployment

Overview

Environment & prerequisites

Usage

Verify tool discovery

​Overview

​Environment & prerequisites

​Usage

​Verify tool discovery

Overview

Environment & prerequisites

Usage

Verify tool discovery