The server has 9 tools with clearly defined input/output schemas using Zod. All tools have descriptions (though of varying quality), and all use verb_noun naming. However, there are significant gaps: (1) Most parameter descriptions are minimal (under 40 chars), violating the baseline of 72 chars average for A+ tools. (2) Output schemas use z.looseObject() which bypasses proper schema validation, this masks what fields clients should expect. (3) Error handling is present but generic; most tools don't provide recovery guidance. (4) No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear semantic differences (e.g., deleteBucket is destructive, listBuckets is read-only). (5) Descriptions lack dependency hints, multi-step guidance, or prerequisites. Tool naming is solid (createBucket, deleteBucket, listFileNames), but descriptions and parameter context are weak. Baseline for A+ tools: 100% have param descriptions; this server is ~30% compliant. Overall definition quality is fair-to-good (C+ range), held down by shallow parameter documentation and missing semantic hints.
Create a new B2 bucket with the specified name and type
Delete a B2 bucket by its ID
Get information about a specific bucket by name
Get a URL for uploading files to a bucket
List all B2 buckets in the account
List files in a bucket
List all versions of files in a bucket
Output schemas use z.looseObject() instead of strictly-typed schemas, bypassing validation and making it impossible for clients to know what fields will be returned. The Zod schemas define expected structure but z.looseObject() allows any fields to pass through.
Parameter descriptions are minimal (under 40 characters for most), far below the A+ baseline of 72 chars average. Examples: 'ID of the bucket to delete' (28 chars), 'Bucket type: allPublic or allPrivate' (36 chars). Descriptions lack context on format, constraints, or multi-step dependencies.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | D | 57 | 2025-06-18+ | v2 |
| 2026-03-09 | F | 0 | - | v1 |
Update a bucket's type (public/private)
Upload a file to B2. Data should be base64-encoded.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear semantic differences. deleteBucket should be marked destructive, listBuckets/listFileNames should be marked readOnly. LLMs cannot determine safe-to-retry vs irreversible operations.
Error handling is generic (code, message, status), but does not provide recovery guidance. Example: 'Error Code: bucket_not_found' tells the agent nothing about what to do next (retry? search for alternatives? check spelling?). Missing pattern:recovery-guide.
uploadFile requires base64-encoded data and returns minimal context. No guidance on file size limits, encoding requirements, or how to chain the result to subsequent operations. Parameter descriptions lack format details (e.g., 'Data must be base64-encoded, max X bytes').
listFileNames and listFileVersions accept optional parameters (startFileName, maxFileCount, prefix) with no documented constraints or defaults. No guidance on pagination or result limits. A+ tools document min/max bounds (e.g., maxFileCount: 1-10000).
Credentials (B2_APPLICATION_KEY_ID, B2_APPLICATION_KEY) are correctly injected server-side, avoiding exposure in tool parameters. This is correct; no issue here.