Read Microsoft 365 attachments with explicit extraction coverage
Hand a selected Microsoft 365 attachment to an approved processor through a fresh download URL, preserving continuation and extraction-coverage evidence.
npx skills add agentskillexchange/skills --skill read-microsoft-365-attachments-explicit-extraction-coverage
Invoke this MCP workflow when an agent must answer a bounded question from a Microsoft 365 attachment using an external document processor. The operator specifies the message, attachment, account and relevant content scope. Return extracted evidence, source identity, processor-native completion and coverage fields, warnings and gaps. This makes partial reading accountable when ordinary attachment browsing would conceal omitted content.
What this skill actually does
The upstream guide specifies list-mail-attachments with id, name, contentType, size and isInline selected, then get-download-url for the chosen attachment’s $value route. Attachment URLs require appropriately configured HTTP mode with --enable-attachment-urls and necessary mail permissions. Give the temporary URL only to an explicitly approved processor that accepts it; Microsoft credentials remain with the integration. Inspect the processor’s real tool schema instead of assuming a universal converter API.
Server-minted URLs are single-use. Mint a fresh ticket for each probe, conversion, continuation or retry, preserving source and account. Pass exact returned continuation values unchanged and retain the complete native result envelope. Exhausted output does not establish full source coverage, especially with page selections, source truncation or OCR omissions. Multi-GET processors need a compatible fresh-ticket strategy.
The scope boundary is one authorized attachment read and evidence handoff. Exclude message classification, moving or sending mail and tenant administration. MCP is central to invocation. The guide is pinned to a development snapshot; verify deployed tool and permission behavior. No converter is supplied, and PDF/OCR fidelity is not guaranteed.