anthropics / anthropics/claude-ai-mcp
M365 connector: sharepoint_upload_file 1 MiB cap removes the only verifiable write path for common files
- Ngôn ngữ chính
- Không có dữ liệu ngôn ngữ
- Star
- 471
- Fork
- 76
- Chỉ số merge pull request
- Không có pull request nào được merge trong 30 ngày
Mô tả
### What happened?
`sharepoint_upload_file` in the Microsoft 365 connector caps content at 1048576 bytes. From the current tool definition, verbatim:
> "Create a new file in a SharePoint document library or OneDrive folder (content up to 1048576 bytes)."
> "Decoded size must not exceed 1048576 bytes."
> "do not split a single binary file across calls — there is no chunked upload."
Microsoft Graph supports simple upload to 4 MB and resumable upload sessions for files up to 250 GB. The cap is the connector's, not Graph's.
**Why this is a correctness problem, not a size inconvenience.**
Below the cap, an agent filing a document gets an item id back from the server. That is proof the file arrived.
Above the cap there is no API path at all, so the only remaining way to file a document is to move it into the local OneDrive sync folder. That path fails silently. If the sync client is not running, the file sits on local disk looking filed, and nothing — not the agent, not the user — reports otherwise. I hit exactly this: the sync client was closed, and files appeared saved when nothing had been uploaded.
So the cap does not make large files inconvenient. It removes the only verifiable write path for them and substitutes one that can report a success it has not achieved.
A scanned PDF routinely exceeds 1 MiB, so this is the common case rather than an edge case.
### What did you expect to happen?
Either of these would close it:
- an upload-session-backed variant of `sharepoint_upload_file`, using Graph's resumable upload sessions
- or a higher cap, matching at least Graph's 4 MB simple-upload limit
If the cap exists to bound response size or request duration, an upload session addresses both directly — that is what it is for.
### Steps to reproduce
1. Connect the Microsoft 365 connector in Claude.ai.
2. Call `sharepoint_upload_file` with any file over 1048576 bytes — a scanned PDF is the ordinary case.
3. The call is rejected on size. There is no chunked or resumable variant to fall back to, as the tool definition itself states.
4. The only remaining route is the local OneDrive sync folder. To see the silent failure: quit the OneDrive sync client, move a file into the synced folder, and observe that it appears filed while nothing has been uploaded. Nothing in the agent's result, and nothing on screen, distinguishes this from a successful upload.
### Area
Tool Discovery / Invocation
### MCP Server (if applicable)
Microsoft 365 (first-party connector in Claude.ai)
### Error messages or logs
```shell
No runtime error — the constraint is stated in the tool definition itself:
"Create a new file in a SharePoint document library or OneDrive folder (content up to 1048576 bytes)."
"Decoded size must not exceed 1048576 bytes."
"do not split a single binary file across calls — there is no chunked upload."
```
### Additional context
Reported to support by email first. The reply confirmed it could not access the tool definitions, could not route the request to a team or individual, and suggested escalating through another channel.
Related Graph documentation: simple upload is supported to 4 MB, and resumable upload sessions to 250 GB. Nothing in Graph imposes the 1 MiB limit.
Hướng dẫn đóng góp
Chưa lập chỉ mục được hướng dẫn đóng góp cho kho mã nguồn này
Đánh giá
Issue này chưa được đánh giá.