Ocean Dock provides 11 read-only session management tools with consistent naming (verb_noun pattern: list_, get_, search_) and basic parameter schemas. However, descriptions are minimal (10-30 chars), parameter descriptions are sparse or missing, and output schemas are entirely undocumented. No error handling guidance, no pagination support despite returning lists, and no tool annotations. The codebase shows tool registration via @server.tool() decorator with input schemas present, but LLM-facing documentation is insufficient for reliable tool selection and chaining.

• Expand all tool descriptions to 50 - 150 characters. Example: 'list_sessions' → 'List all Claude Code sessions, optionally filtered by project path. Returns session ID, timestamp, message counts, and working directory. Supports pagination via limit parameter.'
• Document output schemas for every tool. Define what fields each returns (e.g., list_sessions returns [{session_id, timestamp, user_count, assistant_count, cwd, first_message}, ...]). Use JSON Schema or explicit type hints in docstrings.
• Add parameter descriptions for all inputs. For 'role' in get_session_conversation, clarify: 'Filter by message role: "user" (user messages only), "assistant" (AI responses only), "all" (both, default).'
• Implement pagination with cursor/offset and total_count. E.g., list_sessions should return {sessions: [...], total: N, next_cursor: '...'} to handle large result sets without context overflow.
• Add error handling guidance. Document failure modes: 'If session_id prefix matches multiple sessions, returns error with list of candidates. If session not found, returns "Session not found" with available project filters.'