This MCP server exhibits severe and pervasive quality issues across nearly all 25 tools. While tool names generally follow verb_noun conventions and input schemas are partially present, the implementation fails critical baselines: descriptions are minimal or absent (averaging ~15-25 chars, well below the 194-char baseline for A+ tools), parameter descriptions are sparse or missing entirely, output schemas are not documented, and error handling lacks recovery guidance. The server also mixes Python (scraper microservice) and Go (backend) across tools without clear abstraction, source files span backend/scraper/server.py, backend/pkg/auth/handler.go, and frontend routes, but actual tool definitions are not visible in the provided code excerpts. Authentication/secretmanagement tools (register, verify, login, change-password) expose security concerns: tool descriptions do not acknowledge password handling or encryption, and there is no evidence of secure credential storage in the MCP tool definitions themselves (this may be handled server-side, but the tool interface itself does not declare scope or permission requirements per the pattern:scope-declaration baseline). The chat and community platform binding tools reference SSE streaming and external platform integration but provide no documentation of response structures, pagination, or error recovery paths.

• Expand all tool descriptions to 100-200 characters minimum, explaining WHAT the tool does, WHEN to use it, and what happens (read vs. write vs. destructive). Example for 'chat_send': 'Send a message to an LLM chat session and receive a streamed response. If session_id is omitted, creates a new session. Returns SSE stream of reasoning tokens (optional) and response text.'
• Add descriptions to every parameter. For 'platform' enum parameters, use: 'Platform to bind (must be one of: douban, weread, flomo). Each platform requires separate authentication via QR code login.' For 'session_state_json', add: 'Serialized session state JSON from previous bind/sync. Must include authentication tokens and sync cursors (format: {"auth_token": string, "sync_cursor": integer}).'
• Document output schemas for every tool. Use JSON Schema format. Example for 'login': Returns {"user_id": integer, "email": string, "token": string, "expires_at": ISO8601-date}. Example for 'chat_list': Returns {"sessions": [{"session_id": string, "title": string, "created_at": ISO8601-date, "message_count": integer}], "total": integer}.
• Add pagination to list tools: 'chat_list', 'chat_messages', 'community_data'. Accept 'limit' (1-100, default 20) and 'offset' or 'cursor' parameters. Return 'has_more' boolean or 'next_cursor' so agents can fetch subsequent pages.
• For destructive tools (delete, chat_delete, chat_batch_delete, community_bind_delete), implement a two-step pattern: (1) add a 'dry_run=true' parameter that returns a preview of what will be deleted, (2) require explicit 'confirm=true' in the actual deletion call. Or add a separate 'confirm_delete' tool that returns a confirmation token requiring re-entry in the delete call.