Hacktoberfest 2026：メンテナが10月に向けて印を付けた、オープンで初心者向けの issue。 Hacktoberfest の issue を見る

OneNote: read handwriting, images and SharePoint notebooks

オープン
  • 何か問題がありますか？
#1,284 コメント 0 件 リアクション 0 件 担当者 0 名 GitHub で見る

メンテナーはふだん 1 日以内に返信

まだ誰も着手していません。

評価

難易度
5/5
見積もり時間
1週間以上
初心者へのやさしさ
30/100
issue の種類
機能追加
明瞭さ
明確に書かれている
活発さ
活発
技術スタック
typescript
領域
ai, backend-api-design

調査の方向性

Start with intl/microsoft-onenote.json and connectors/binary-response.util.ts to understand the existing content and binary response paths. The proposal also requires tracing the connector layer and response-transform.util.ts opt-in pattern, then checking Microsoft Graph permissions and calibrating InkML coordinates on real pages. Done means resource retrieval, SharePoint and group notebook access, and optional page images with ink work within the stated response limits.

索引モデルが issue の本文から書いたものです。

説明

enhancement
Problem

The OneNote adapter (intl/microsoft-onenote.json) reads a page through GET /me/onenote/pages/{id}/content, which returns HTML without ink. Many OneNote users write by hand with a pen (meeting notes, sketches, calculations), so the AI sees nothing of those pages. Embedded pictures and PDF printouts reach the AI only as <img> links to Graph that need the user's sign-in, so it cannot read them either.

Two smaller gaps in the same adapter:

  • Notebooks stored on SharePoint team sites (/sites/{id}/onenote, /groups/{id}/onenote) are not reachable; only /me/onenote is.
  • There is no tool to fetch a single page resource (image or printout).
Proposed Solution
  1. microsoft_onenote_get_resource (adapter JSON only): GET /me/onenote/resources/{id}/$value with the existing binary handling (connectors/binary-response.util.ts), so an image comes back as an MCP image content block that the model can read.
  2. SharePoint and group notebooks (adapter JSON only): list notebooks, sections and pages under /sites/{site-id}/onenote and /groups/{group-id}/onenote. Check which delegated permission Graph needs for these (likely Notes.Read.All / Notes.ReadWrite.All, admin consent) and document it in the adapter instructions.
  3. microsoft_onenote_get_page_image (engine change): render a page as PNG including ink.
    • Fetch content?includeinkML=true (HTML plus InkML).
    • Parse the InkML traces (fast-xml-parser, already a dependency) and draw each trace as an SVG path with its brush colour and width, scaled to page coordinates.
    • Fetch the page's images with the user's token and place them at their absolute positions.
    • Rasterise with @resvg/resvg-js (MPL-2.0, no browser needed); split tall pages into tiles of about 1,500 px on the long edge.
    • Return the typed text as text (as today) plus the image(s). Typed text is not drawn into the PNG; that would need a headless browser in the image.
    • Opt-in per tool through a config key (e.g. "render": "onenote-page"), following the opt-in pattern of decode / images in response-transform.util.ts, but run in the connector layer because it needs further authenticated Graph calls.
Alternatives Considered
  • Handwriting recognition on our side: not needed; current models read handwriting from images well, and Graph exposes no recognised text for ink.
  • Rendering the full page in headless Chromium: higher fidelity for typed text, but adds about 300 MB to the image and a lot of memory per render on the shared Cloud.
  • Returning raw InkML: the model cannot use stroke coordinates meaningfully.
Area

Connectors / adapters (Microsoft OneNote), REST engine.

Additional Context
  • Cloud and self-hosted behave the same; the new dependency ships in the backend image for both.
  • Size limits: the 10 MB tool response cap and the per-tool binary.maxBytes apply; tiles keep each image well below them.
  • InkML coordinate units need calibrating against real pages with pen input before release.
主要言語
TypeScript
スター
856
フォーク
82
平均マージ
7時間 49分
マージ済み PR（30日）
302

環境構築

このプロジェクトには開発コンテナ、Dockerfile、コントリビューションガイドがありません。まず README を読み、一般的な手順ははじめてのコントリビューションガイドを参照してください。

はじめの一歩

  1. issue を最後まで読み、次にプロジェクトのコントリビューションガイドを読みます。
  2. 着手することを issue にコメントします — 二人が同じ作業をするのを防げます。
  3. リポジトリをフォークし、ブランチを切って変更します。
  4. issue 番号を参照したプルリクエストを送ります。

HelpCode-ai/anythingmcp のほかの issue

HelpCode-ai/anythingmcp の issue をすべて見る

似ている issue

TypeScript の issue をもっと見る

新しい issue をメールで受け取る

初心者向けの GitHub issue を短くまとめたダイジェスト。