Repository navigation
Git Vertex MCP v0.5.3
·
135 commits
to 5793b473078a9e0258c06e9be64dffa50e974cda
since this release
Git Vertex MCP v0.5.3
Fixed
git_bisectnever told you it had found the answer in a way you could rely on. git announces it in prose, and the wording depends on the git you run: 2.39 says<sha> is the first bad commit, 2.55 says<sha> is the first 'bad' commit— quoted, because bisect terms are configurable. So a client keying on either string works until the day git is upgraded, with no change on its side. The tool now ends its output with a stablefirst bad commit: <sha> — <subject>line, derived fromrev-list --bisect-vars(shell assignments: never translated, never quoted). The tool's own test suite was the first victim — it looped through every step of a bisect that converged perfectly and concluded it had not converged.- A converged session also claimed a "currently checked out" commit — which is the last commit tested, not the culprit, and one line above the answer. It is now one or the other: where to test next, or the answer.
- The server announced the wrong version to every client.
VERSIONwas a constant kept in step by hand and it had drifted:serverInfoand the startup line said0.4.0while the package was0.5.2. It is read frompackage.jsonnow, and a test compares the two.
Local MCP server exposing your Git repositories to AI agents (Claude Code, Cursor, Copilot…). Read-mostly, no cloud.
npx git-vertex-mcp
# or
npm i -g git-vertex-mcp && gv-mcp
🔗 npm