Rainkeeper demonstrates solid tool design with consistent naming (verb_noun pattern), complete input schemas with types, and clear descriptions for most tools. All 19 tools follow the verb_noun convention (list_, get_, create_, update_, delete_, search_, bulk_, rename_, move_). Descriptions average ~120 chars and explain WHAT the tool does and WHEN to use it. However, output schemas are not documented, LLMs cannot predict response structure for chaining. Error handling is present in the client layer but not surfaced in tool descriptions. Bulk operations lack per-item success/failure granularity. No tool annotations (readOnlyHint, destructiveHint) despite clear risk stratification (READ_ONLY, WRITE, DESTRUCTIVE). Parameter descriptions are present but some lack format/constraint details (e.g., 'sort' accepts '-created, title, -title, domain' but no enum). Overall: good naming and schema coverage, but missing output documentation and annotations.
Permanently delete multiple bookmarks. ids must be non-empty.
Flag multiple bookmarks as important (starred). ids must be non-empty.
Move multiple bookmarks to a target collection. ids must be non-empty and all from source_collection_id — ids from other collections are silently skipped by the API.
Apply tags to multiple bookmarks. Merges with existing tags. WARNING: passing tags=[] will remove all tags from the targeted bookmarks.
Create a new collection. Pass parent_id to create a sub-collection.
Save a new bookmark. collection_id -1 = Unsorted. Tags is a list of strings. Pass title='' to explicitly clear the auto-detected title.
Output schemas not documented. LLMs cannot predict response structure (fields, types, nested objects) for downstream tool chaining. E.g., list_collections returns 'items' array but structure of each item is undocumented.
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). Tools are risk-stratified in metadata (READ_ONLY, WRITE, DESTRUCTIVE) but not exposed to LLM via MCP annotations. Agents cannot distinguish safe reads from destructive deletes.
Bulk operations lack per-item success/failure reporting. bulk_delete_raindrops, bulk_move_raindrops, bulk_tag_raindrops return a single response, if one item fails, the agent cannot determine which succeeded and which failed, forcing a full retry.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 63 | 2026-07-28+ | v2 |
Delete a collection. Contents move to Unsorted.
Delete a bookmark permanently.
Delete one or more tags from all bookmarks.
Get details of a single collection by id.
Get a single bookmark by id.
List all Raindrop.io collections (root and sub-collections), with id, title, and parent_id.
List bookmarks in a collection. collection_id 0 = All, -1 = Unsorted. sort: -created, title, -title, domain. Maximum per_page is 50.
List all tags. Pass collection_id to scope to a specific collection.
Move a single bookmark to a different collection. Explicit tool for clarity.
Rename a tag across all bookmarks.
Search bookmarks using Raindrop search syntax. collection_id 0 = search all. Supports #tag, word, "phrase", type:article, etc. Maximum per_page is 50.
Update a collection's title, visibility, or colour. color is a hex string e.g. '#ff0000'.
Update a bookmark. Pass collection_id to move it — uses the correct nested API format.
Enum constraints not declared in schemas. 'sort' parameter accepts '-created, title, -title, domain' but is a free-form string, not an enum. LLMs may hallucinate invalid sort values. Same for search query syntax, undocumented.
Error handling not surfaced in tool descriptions. Client layer has clear error messages (401 auth, 404 not found, 429 rate limit, 500 server error) but tool descriptions do not mention recovery steps. LLMs cannot self-correct on failures.