Model Context Protocol server for Memos - a self-hosted note-taking service
Strong foundation with 17 well-named tools, comprehensive schemas, and detailed parameter descriptions. Tool names follow verb_noun convention (list_memos, create_memo, update_memo, delete_memo, etc.). Most parameters include type constraints and descriptions. However, several tools lack output schema documentation, error handling guidance is minimal, and some descriptions could be more LLM-optimized. Pagination support is present (pageSize, pageToken in list_memos) but not consistently applied. Tool composition is sound, each tool has a single responsibility and output IDs chain to downstream tools.
Mark the current review batch as completed. Call this after the user has finished reviewing all memos from get_review_memos.
Create a new memo with markdown content.
Delete a memo permanently.
Delete a resource permanently.
Get a single memo by its numeric ID or UID string (the ID from the memo URL).
Get memos created on this day in previous years, grouped by year. Great for revisiting past thoughts and memories.
Output schemas not documented. Tools return data but LLMs cannot see field structure (e.g., list_memos returns memos array, but field names/types are not declared in tool definition). This forces LLMs to infer structure and risks misuse of returned data.
Error handling guidance missing. Tools have no documented error cases or recovery paths. E.g., delete_memo does not explain what happens if memo not found, or if user lacks permission. LLMs cannot self-correct on failures.
Destructive operations lack confirmation step. delete_memo, delete_resource, and rename_tag are irreversible but have no dry-run or confirmation mechanism. Agents can accidentally destroy data.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 60 | 2026-07-28+ | v2 |
Get today's batch of memos for spaced-repetition review (回顾). Returns a daily batch selected by the review algorithm.
List relations of a memo. Returns edges with direction (from → to) and type. - depth=1 (default): direct relations only. - depth=2+: recursively expand related memos to build a relation graph. Each edge shows {from, to, type}. Use get_memo to inspect specific nodes.
Search and list memos. All filters are optional and combined with AND logic.
List all resources (attachments) owned by the current user.
List tags with usage counts. Returns {name, count, hasChildren, pinned, emoji} for each tag. - No args: top-level tags only. - recursive=true: every tag (flat list, good for recommending tags). - parent="x": direct children of tag "x". - parent="x" + recursive=true: all descendants under "x". - pinnedOnly=true: only user-pinned tags. Tags use "/" for hierarchy (e.g. "project/work/backend").
Rename a tag across ALL memos. This is a destructive operation that modifies memo content globally — use with caution.
Add or remove REFERENCE relations for a memo. - action="add": add new references without affecting existing ones. - action="remove": remove specific references. Note: COMMENT relations are managed by the system and cannot be modified here.
Update an existing memo. Only the fields you provide will be modified.
Update review preferences: how many memos per batch, and which tags to include or exclude.
Update a tag's pinned status or emoji. Pinned tags are highlighted in the UI for quick access.
Upload a file as a resource. The file content must be base64 encoded. Optionally link it to a memo.
Inconsistent pagination. list_memos supports pageSize and pageToken, but list_tags, list_resources, and list_memo_relations do not. Large result sets could exhaust context window.
Dual ID acceptance (numeric vs UID) not consistently documented. get_memo, update_memo, delete_memo accept both numeric ID and UID string, but descriptions do not clearly explain when to use each or how LLM should choose. Ambiguity invites wrong ID format.