MCP server for RomM — browse and manage your retro game library with AI
mcp-romm defines 40 tools with consistent naming (verb_noun: romm_status, romm_platforms, etc.) and reasonable descriptions (most 40-150 chars). All tools have parameter schemas with type definitions visible in smoke_test.py. However, critical gaps reduce overall quality: (1) Parameter descriptions are minimal or absent in many tools, e.g., romm_library_items has 'limit' and 'offset' described generically without context about what gets filtered; (2) Output schemas are entirely undocumented, no tool explicitly states what fields it returns, forcing LLMs to guess response structure; (3) Error handling guidance is absent, no tool describes what to do on 404, timeout, or auth failure; (4) The 'filter_rules' parameter in smart collection tools is typed as 'object' with no documentation of valid keys/values, inviting hallucinated filters; (5) Write tools (romm_set_status, romm_add_to_collection) lack explicit guidance on side effects and idempotency. Naming is strong across the board, all tools start with action verbs and are unambiguous. Descriptions are present but generic: most restate the tool name rather than adding context (e.g., 'Add a note to a ROM' vs 'Attach a private note to a ROM to track personal observations, bugs, or play history'). The server demonstrates competent API wrapping but falls short of LLM-optimized tool design.
Recent play activity feed (RomM 5.0+)
Add a note to a ROM
Add ROMs to a collection
List ROMs in a specific collection
List user-curated collections
Create a new collection
Create a smart collection (saved filter)
Output schemas completely undocumented. No tool explicitly declares what fields it returns. This forces LLMs to guess response structure, leading to failed field extraction and unnecessary retry logic.
Parameter 'filter_rules' in romm_create_smart_collection and romm_update_smart_collection is typed as 'object' with no documentation of valid keys, values, or structure. LLMs cannot construct valid filters without explicit schema.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 66 | <=2025-11-25 | v2 |
Delete a collection (permanent)
Delete a note (permanent)
Delete a play session (permanent)
Delete a smart collection (permanent)
List registered devices
Add or remove a ROM from your favorites
Available filter values (genres, regions, etc.)
List BIOS/firmware files per platform
Full ROM detail (metadata, saves, user status)
Browse ROMs with filtering and pagination
Record a play session (RomM 5.0+)
Search metadata providers for ROM matches
List platforms with ROM counts and sizes
List recorded play sessions (RomM 5.0+)
Recently added or updated ROMs
Remove ROMs from a collection
View notes on a ROM
List save files by ROM or platform
Trigger a background library rescan
Search ROMs by name
Identify a ROM by file hash (MD5, SHA1, CRC, or RetroAchievements)
Set play status, backlog, now-playing, rating, completion
Smart collection rules + matching ROMs
List auto-generated smart collections
Library-wide statistics
Check server configuration and reachability
Task registry + running task status
Edit an existing note
Edit a smart collection
Browse by user status (now playing, backlog, etc.)
List ROMs in a virtual collection
Automatic metadata groupings (RomM 5.0+)
Authenticated account, role, and permissions
Error handling guidance is absent. Tools do not document what happens on 404 (resource not found), 401 (auth failure), timeout, or server error, and do not suggest recovery steps. LLMs cannot self-correct or plan fallbacks.
Write tools (romm_set_status, romm_favorite, romm_add_note, romm_log_play_session, etc.) lack explicit idempotency guidance. Agents retrying on ambiguous failures need to know: is it safe to call twice with the same input?
Destructive operations (romm_delete_note, romm_delete_play_session, romm_delete_collection, romm_delete_smart_collection) lack confirmation or dry-run support. Agents can permanently delete user data without explicit approval.
Many parameter descriptions are generic or missing context. E.g., romm_library_items has 'limit' and 'offset' described as 'Number of items per page' and 'Pagination offset' without saying what data gets paginated or typical range. Baseline recommends 50-200 chars with actionable constraints.
Tool descriptions often restate the tool name rather than adding actionable context. E.g., 'Add or remove a ROM from your favorites' vs 'Mark a ROM as a favorite for quick access in your library, or remove it from favorites. Favorites appear at the top of the library and sync across sessions.'