research-mcp defines 14 tools with consistent naming (verb_noun pattern: register, get_state, advance, log_artifact, etc.) and reasonable descriptions (50-150 chars). All tools have input schemas with typed parameters and enums where appropriate (kind, status). However, parameter descriptions are sparse or missing entirely, most params lack explanatory text beyond the name. Output schemas are not documented, forcing LLMs to infer response structure. Error handling is minimal; no recovery guidance or actionable error messages. The server follows a coherent domain model (research projects, experiments, tables) but lacks the depth of documentation and error resilience expected in production tools.

• Add 50 - 150 character descriptions to every parameter. For 'params' in save_experiment, write: 'JSON-encoded hyperparameters (e.g., {"learning_rate": 0.001, "batch_size": 32}). Must be valid JSON string.'
• Document output schemas for all tools. E.g., get_state should return {state: string, validNextStates: string[], transitions: {state: string, reason?: string}[]}.
• Add error handling guidance. E.g., get_state: 'Returns 404 if project not found. Try list_projects() to verify the slug exists.'
• Add optional limit and offset parameters to list_projects, list_experiments, and list_tables. Document default limit (e.g., 20) and max (e.g., 100).
• Add a confirm_delete parameter (default false) to delete_experiment and delete_table, or create separate preview_delete tools that show what would be deleted before committing.