Repository navigation
UVCS MCP 1.4.0
UVCS MCP 1.4.0
UVCS MCP 1.4.0 lets agents understand repository history, review branches, preview merges, check in only the files of a task, and lock Unity assets before editing them. It ships an agent skill that teaches Claude Code, Codex, Cursor, OpenCode, and other SKILL.md agents to use these tools. It also fixes three issues that affect every earlier release, so upgrading is recommended.
Fixes
uvcs_diff_fileno longer starts the GUI diff tool. In 1.3.0 and earlier it rancm diff <path>, which opened a diff window on the user's desktop and hung until the read timeout. It now returns a unified diff of the loaded revision against the file on disk. The result has adifffield instead of the rawcmoutput.uvcs_mergeno longer leaves a conflicting merge in progress. Prepare runs the merge preview and refuses merges with conflicts (MERGE_HAS_CONFLICTS); confirm checks again.- Checkin tokens bind the size and modification time of the files. Editing an already changed file after prepare used to go into the checkin unreviewed; now the confirm is refused with
WORKSPACE_CHANGED_SINCE_PREPARE. uvcs_changeset_analyticssends dates as ISOyyyy-MM-dd; the previous form was rejected on some locales.
History, review, and merge preview
All read-only, in every safety profile:
uvcs_changeset_diff: what changed in a changeset, label, shelve, or branch, or between two changesets or labels, with Unity.metapairing warnings.uvcs_revision_diff: the line diff of one file for a changeset, between revisions, or against the workspace copy.uvcs_file_historyanduvcs_annotate: who changed a file or a line, and when.uvcs_branch_listanduvcs_label_list: find branches and labels by name, owner, date, parent, or branch.uvcs_merge_preview: files a merge would bring and items that would conflict, into the workspace or into any branch on the server.
Checkin by path and exclusive checkout
uvcs_checkin_prepareacceptspaths. Selected Unity assets bring their.metafiles along, assets whose.metais still private are refused (UNITY_META_NOT_ADDED), and new parent folders a change needs are listed asdependencies.uvcs_checkoutchecks files out and takes the exclusive lock when the server'slock.confcovers them. Files locked by someone else are refused.uvcs_release_checkoutreleases checkouts of unchanged files; modified files keep their checkout and edits.uvcs_locksacceptspathsand ascope(all,user,workspace).
Agent skill
uvcs-mcp skill installcopies theuvcs-version-controlskill into Claude Code (.claude/skills), the shared.agents/skillsfolder (Codex, Cursor, OpenCode, Gemini CLI), and other agent folders, for the user or a project.npm run pack:skillvalidates the skill and builds a marketplace ZIP.
Safety
- Item paths containing
*or?are rejected, becausecmexpands wildcards itself. zod4.6.7 and transitive dependency updates;npm auditreports zero known vulnerabilities.
Quality gate
- 169 automated tests (one POSIX-only test skipped on Windows), ESLint, syntax validation, and mutation checks of the key safety guards.
- Fake single-workspace and fleet MCP smoke flows, including checkout, release checkout, path-scoped checkin, and the history tools.
npm run smoke:packinstalls the packed tarball, starts the installed server, and checks that the skill is shipped.- Live validation with
cm10.0.16.6656on a self-hosted server: the Plastic SCM smoke and a 37-check write scenario on a disposable repository (path-scoped checkins,.metapairing, moves into new folders, deletes, checkout and release, a real merge conflict), and the read-only history tools on a production repository. - Not covered live: exclusive locking through a
lock.confrule, and the new commands oncm11.x clients.
Upgrade
npx -y @proanima/uvcs-mcp@1.4.0 init --client=claude-code --workspace="D:/Repositories/YourWorkspace" --safety=guarded
npx -y @proanima/uvcs-mcp@1.4.0 doctor --workspace="D:/Repositories/YourWorkspace"
npx -y @proanima/uvcs-mcp@1.4.0 skill installRestart the MCP client after upgrading. Clients that parsed the stdout of uvcs_diff_file should read diff instead. See the changelog, the agent skill guide, and the security model.