RpgLedger demonstrates mixed quality across its 13 tools. Strengths: all tools have names starting with action verbs (list_, get_, create_, update_, upsert_, delete_, roll_); schemas are present for all tools with typed parameters; many parameters have descriptions. Weaknesses: (1) Descriptions vary significantly in quality, some are detailed and multi-line (create_campaign, create_character, update_character, upsert_location, delete_location), others are minimal or generic (list_campaigns 'List all available campaigns', roll_dice description is English but terse); (2) Mixed language use, descriptions for some tools are in Polish (list_locations, create_campaign, create_character, update_character, upsert_location, delete_location, list_factions, upsert_faction, faction_rep_add), others in English, which reduces clarity for English-speaking agents; (3) Output schemas are not documented, callers cannot see what structure list_campaigns, get_campaign, or get_character return; (4) No error handling guidance in descriptions, tools like delete_location (destructive) do not warn of consequences or recovery paths; (5) Parameters lack constraints, e.g., campaign_id and char_id accept any string, with no format validation guidance; (6) Roll_dice includes optional logging but no guidance on when or why to use it. Average tool description length is ~150 chars (within rubric baseline of 194), but variance is high and Polish descriptions reduce usefulness. Per-tool assessment: naming is consistently strong (100 across all tools), descriptions range 20 - 90 depending on language and detail, schemas are complete but undocumented return types.

• Add output schema documentation to every tool. For list_campaigns, document: returns List[{id: string, name: string}]. For get_campaign, document: returns {id, name, day, location, characters: [], locations: [], factions: [], ...}. For create_campaign, document return structure. This enables LLMs to chain calls correctly.
• Standardize all descriptions to English or provide consistent translations. Pick one language for the MCP server and translate Polish descriptions (list_locations, create_campaign, etc.) to English. Mixed languages confuse LLM prompt engineers and reduce clarity.
• Expand minimal descriptions to 50 - 150 characters. Replace 'List all available campaigns' with 'Retrieve a list of all RPG campaigns available in this ledger. Use this to discover campaigns before fetching details with get_campaign.' Include WHEN to call it and what comes next.
• Add recovery guidance to error-prone tools. For delete_location, add: 'WARNING: Deletion is irreversible. If you need to restore a location, you must recreate it with upsert_location or restore from a backup campaign using create_campaign(from_campaign_id=...).' For create_campaign, add: 'If campaign already exists, set overwrite=true or list existing campaigns first with list_campaigns.'
• Declare format constraints for all string IDs. Document campaign_id as 'Kebab-case alphanumeric string (a-z, 0-9, dash), max 64 chars (e.g., smolenia-na-wieczne-tymczasem).' Document char_id similarly. This prevents LLM hallucination of invalid IDs.
• Add pagination support to list_* tools. Accept limit (1 - 100, default 20) and offset parameters. Return {items: [...], total: N, limit, offset}. Document: 'Returns up to 20 campaigns per call. Use offset to paginate through large campaign lists.'