This server has severe definition quality issues. All 14 tools lack visible input schemas in the source code, only tool names and descriptions are provided in the hook scripts. No JSON Schema definitions, parameter types, or constraints are documented. Descriptions are present but generic (50-150 chars), lacking actionable guidance on when/why to use each tool or what parameters they accept. The tools appear to be memory-management utilities, but without schemas, LLMs cannot determine valid inputs or expected outputs. Tool composition is reasonable (each has a single responsibility), but the lack of formal definitions makes this unsuitable for production agent use.

• Create a formal MCP tool manifest (e.g., in .mcp.json or a tools.json file) that explicitly registers all 14 tools with complete JSON Schema input definitions, parameter types, descriptions, and constraints.
• For each tool, document the output schema with field names, types, and descriptions. E.g., mem_search should return {observations: [{id, content, timestamp, topic_key}], total_count, next_cursor}.
• Expand tool descriptions to 100-200 characters, including: what the tool does, when to call it, what parameters it requires, and what it returns. Example: 'mem_save: Persist a decision, bug fix, or discovery to memory. Call immediately after resolving an issue. Requires: content (string), topic_key (optional). Returns: observation_id, timestamp.'
• Add parameter descriptions for all inputs. E.g., mem_save should document: content (string, max 2000 chars, the observation to save), topic_key (string, optional, category for organizing memory), session_id (string, optional, session context).
• Document error cases and recovery paths. E.g., 'mem_delete: If observation not found, returns 404 with message "Observation {id} not found. Call mem_search to find valid IDs." If permission denied, returns 403 with guidance to check session context.'