You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
get_file_contents fails on every file over 1 MB with "unsupported content encoding: none" #3417
Since v1.10.0, get_file_contents returns an error for any file of 1 MB or more instead of the "too large to display" resource link it returned before. The Contents API sends these files with "content": "" and "encoding": "none", and repositoryContentBytes (pkg/github/repositories_helper.go:220) calls GetContent() on that, which go-github rejects. This happens before the large-file branch at pkg/github/repositories.go:1118 is reached.
The existing "large file returns ResourceLink" tests mock the Contents API response without content/encoding fields, so they don't hit this.
Affected version
GitHub MCP Server
Version: 1.14.0
Commit: f10e4e1f923d46b86f2e80e849aa74084c847184
Build Date: 2026-10-02T17:00:10Z
(5,179,756 bytes; any file of 1 MB or more fails the same way, text or binary.)
Expected vs actual behavior
Expected (what v1.9.0 returns):
File drivers/gpu/drm/amd/include/asic_reg/nbio/nbio_7_4_sh_mask.h is too large to display (5179756 bytes). Use the download URL to fetch the content: https://raw.githubusercontent.com/torvalds/linux/a90ee43.../nbio_7_4_sh_mask.h (SHA: 07f04b2b...)
plus a resource_link.
Actual (v1.14.0, isError: true):
failed to inspect repository file: failed to decode file content: unsupported content encoding: none, this may occur when file size > 1 MB, if that is the case consider using DownloadContents
Logs
Contents API response for that file: {"type":"file","size":5179756,"encoding":"none","content":""}.
I have a small fix with a regression test ready and can open a PR: treat encoding: "none" as "content not returned", so the existing large-file path runs again.
Activity