Read-only remote MCP server that lets an AI assistant search and read Microsoft Loop content via the Graph API, with per-user permission trimming.
Six read-only tools with clear, well-structured descriptions (avg 180 chars, all >20 chars). All tools have proper verb-noun naming (loop_search, loop_get_page, etc.). Input schemas present with types and descriptions for all parameters. Numeric parameters include min/max constraints (top: 1-25, query maxLength: 500). Output schemas documented in descriptions. Security model is explicit (permission-trimmed, per-user handles, audit logging). Main gaps: no tool annotations (readOnlyHint missing despite all being read-only), no error recovery guidance in descriptions, and parameter descriptions could be more prescriptive about format/constraints. Composition is sound, tools chain well (search→get_page via ref). No destructive operations, so confirmation patterns not applicable.
Find Microsoft Loop pages that look like meeting notes (best-effort, by name and folder path — Loop meeting notes are not a distinct type). Optionally narrow with a query. Each result includes a `ref` for loop_get_page. Read-only.
Read a Microsoft Loop page/component as HTML. First call loop_search or loop_list_components, then pass the `ref` string from the result you want. Each ref is per-user and time-limited; only items YOU discovered can be read. Read-only.
List Microsoft Loop component files (.loop) stored in the calling user's OneDrive (Teams/Outlook Loop components). Read-only.
List your most recently modified Microsoft Loop pages (permission-trimmed). Each page includes a `ref` for loop_get_page. Read-only. Note: this reflects recently *modified* pages you can access, which approximates Loop's Recent list.
List the Microsoft Loop workspaces/locations you have access to, grouped from your own permission-trimmed search results. Each workspace includes its pages with a `ref` you can pass to loop_get_page. Optionally narrow with a query. Read-only. Note: a workspace only appears if it has at least one page discoverable by your search.
Missing tool annotations. All six tools are read-only but lack readOnlyHint annotation in schema. This prevents clients from optimizing caching, retry logic, or UI presentation.
Error handling descriptions lack recovery guidance. Tools do not document what LLM should do on failure (e.g., 'If search returns 0 results, try loop_list_recent or loop_list_workspaces'). Descriptions state 'Read-only' but not failure modes.
Parameter 'query' in loop_search and loop_find_meeting_notes lacks format/constraint details in description. Descriptions say 'Search query string' but do not explain sanitization, operator restrictions, or what happens with special characters.
Inferred effective spec: 2026-07-28+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 69 | 2026-07-28+ | v2 |
Full-text search across Microsoft Loop pages and components. Results are permission-trimmed to the calling user. Read-only. Always search before reading a page.
loop_list_components has empty input schema ({}). While correct for a no-parameter tool, description could hint at what it returns (e.g., 'Returns .loop files from your OneDrive; use ref with loop_get_page').
Output schemas documented in descriptions but not in formal JSON Schema. Responses include 'results', 'count', 'note' fields, but no explicit outputSchema property visible in tool registration.