Lexiang Knowledge Base MCP Connector

Problem

This skill targets MCP workflows that need to connect to Lexiang Knowledge Base. Typical issues include deciding whether an action should read, write, upload a file, or edit a Block when users provide a lexiangla.com link, space_id, or entry_id; retrying after 401 or token expiry; guessing write targets from names; generating links with the wrong domain or company_from; and answering questions about existing knowledge without searching first.

How It Works

• Trigger and routing: When the request mentions Lexiang, knowledge base, lexiang, a Lexiang URL, or IDs, the skill routes to the relevant reference file, such as references/setup.md, references/search.md, references/writer.md, references/blocks.md, or references/files.md, instead of loading everything at once.
• Search first: For prompts involving documents, notes, earlier drafts, or writing style, it calls search_kb_search or search_kb_embedding_search before producing analysis.
• Read/write separation: /pages/xxx is used for reading or Block-level editing; new documents under /spaces/xxx use the writer path; Markdown documents are handled separately from PDF, Word, and image files.
• Write confirmation: Writes require an explicit user-provided URL, ID, or confirmed name; it must not traverse knowledge bases and choose a target on its own.
• Link generation: Use company.company_domain from whoami(); for top-level domains, include the source via ?company_from=xxx.
• Batch protection: When file or entry counts exceed 20, work in batches of no more than 20 and tell the user the count and strategy first.

Boundaries and Notes

This skill assumes MCP is connected and credentials are valid; on 401 or disconnection, stop retrying and follow setup or refresh. It fits search, read, create, edit, file transfer, and meeting recording import tasks. It is not meant for batch writes without confirmation, nor for analyzing existing Lexiang content without first retrieving evidence.