mcp-lark

mcp-lark connects an agent to Feishu (飞书) / Lark through Lark's official OpenAPI MCP server, letting it manage user information, chats, emails, cloud documents, multidimensional tables (Bitable), tasks, and calendars. It solves the problem of driving the Lark/Feishu suite from natural-language workflows without hand-writing OpenAPI calls, by routing everything through remote MCP endpoints that you register on the Lark MCP Configuration Platform.

The skill is configured through environment variables, prioritizing a workspace `.env` file and falling back to system environment variables, and prompting the user for values when nothing is set. Multiple MCP service URLs and their usage scenarios are declared in a single `LARK_MCP_SERVERS` variable (for example separating chats-and-mails from cloud-documents), so the agent can pick the right endpoint per task. Tool discovery and invocation are done with `mcporter`: listing all tools and their parameters, fetching a specific tool's JSON schema with `jq`, and calling a named tool with typed parameters. For compatibility the skill recommends running commands via `npx -y mcporter` rather than a globally installed binary, and it bundles a reference document describing message-create payloads.

This skill suits teams that already run on Lark/Feishu and want an AI agent to read and act across their collaboration stack — sending messages, working with cloud docs and Bitable, and managing tasks and calendar events. It is most useful once the MCP URL has been obtained from the configuration platform and stored in the environment, after which the agent can enumerate available tools and call them on demand.