Skip to content

fix(skill): register the MCP server at user scope; give typed commands a 15 s catalog budget - #152

Merged
piyushsarin-sib merged 1 commit into
mainfrom
fix/skill-mcp-user-scope-timeout
Oct 23, 2026
Merged

piyushsarin-sib merged 1 commit into
mainfrom
fix/skill-mcp-user-scope-timeout

Conversation

@piyushsarin-sib

Copy link
Copy Markdown
Collaborator

Summary

Two fixes found while testing the 2.6.0 release candidate. 2.6.0 is not published yet, so they ride in the same release: no new changeset, the existing skill-install-from-catalog entry is extended.

What it fixes

  • --mcp registered the server for one folder only. claude mcp add defaults to local scope, so the Brevo MCP server appeared only in the folder where brevo skill install --mcp ran. gemini mcp add defaults to project scope and is refused in an untrusted folder ("Cannot modify settings in an untrusted workspace"), so registration failed. Both now get --scope user. Codex has one global config and is unchanged.
  • A slow GitHub redirect read as "not reachable". Every catalog fetch gave up after 2 s, including the ones behind a command the user typed. release download redirects are sometimes slower than that (one took over a minute), so skill list and skill install could fail and then work on retry. Typed commands now wait 15 s. The notice printed after other commands keeps the 2 s budget, so it still never holds a command up.

Checks

  • yarn lint, yarn build, full jest suite (1,973 tests) pass.
  • Run by hand with this build and the real claude 2.1.309, codex 0.162.1 and gemini 0.65.0, in an empty HOME and an untrusted folder: skill install --mcp --agent claude,codex,gemini registers the server for all three. claude mcp list shows it from a different folder, and ~/.gemini/settings.json holds it. Running it again reports "already exists" for claude (with the command to run by hand) and registers cleanly for gemini and codex.
  • Gemini CLI itself shows user-level servers as disabled while the current folder is untrusted. That is Gemini's own trust setting, not something this CLI can change.

🤖 Generated with Claude Code

…s a 15 s catalog budget

claude and gemini 'mcp add' default to a per-project scope: Claude Code registered the
server for the current folder only and Gemini CLI refused it in an untrusted folder. Both
now get --scope user.

The catalog fetch gave up after 2 s for every command, so a slow GitHub release redirect
read as 'not reachable'. Commands the user types now wait 15 s; the notice after other
commands keeps the 2 s budget.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
@sonarqubecloud

Copy link
Copy Markdown

@piyushsarin-sib
piyushsarin-sib merged commit 36c1c3f into main Oct 23, 2026
6 checks passed
@piyushsarin-sib
piyushsarin-sib deleted the fix/skill-mcp-user-scope-timeout branch October 23, 2026 10:57
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants