MCP server for Discogs - provides tools to interact with the Discogs music database and marketplace API
Scoring was not performed
Tool descriptions are uniformly terse (10-60 characters). Most fail the pattern:tool-description requirement of 10-1024 chars with clear WHAT/WHEN/WHY context. Examples: 'The Listing resource allows you to view Marketplace listings' (60 chars) and 'Edit a marketplace order' (24 chars) lack guidance on when to call, what the tool returns, or dependencies.
Parameter descriptions are minimal or absent. Schema declares parameter names and types (good), but descriptions lack constraint details. Example: 'condition' parameter in create_marketplace_listing has description 'Condition of the item' but does not specify valid values (NM, VG+, VG, etc.). LLMs cannot infer constraints from type alone.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 17 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 37 | - | v1 |
Output schemas are not documented in the tool definitions. source code does not include example responses or response field documentation. Agents do not know what fields to expect, preventing downstream tool chaining optimization (e.g., will get_release return a label_id to pass to get_label?).
No error handling guidance. The code uses createDiscogsError() for error wrapping, but tool descriptions do not document what errors are possible or how to recover. For example, search_database does not state what happens if the search returns 0 results or if the API rate limit is exceeded.
Destructive operations lack confirmation steps. delete_marketplace_listing and delete_release_user_rating accept minimal parameters and have no dry-run or confirmation mechanism. An LLM error could trigger unintended deletion without safeguards.
Pagination parameters (page, per_page) appear in many tools but are not consistently documented. No mention of maximum page size, default limits, or behavior when page exceeds available results. This violates pattern:paginated-result which requires LLMs to understand result boundaries.
Tool naming uses 'get_' for read operations (good), but edit_marketplace_order diverges from the pattern by using 'edit' instead of 'update'. This inconsistency forces LLMs to reason about whether 'edit' and 'update' are interchangeable or if they modify different fields.
No validation of enum-like parameters. Fields like 'condition', 'status', 'sleeve_condition', 'sort', 'sort_order' accept strings but do not declare valid options as JSON Schema enums. LLMs will hallucinate invalid values (e.g., 'Like New' instead of 'NM', 'pending' instead of valid Discogs statuses).