A Node.js implementation of a Git repository browser using the Model Context Protocol (MCP)
27 tools with complete input schemas and descriptions, but quality is inconsistent. Naming follows verb_noun convention (git_*) appropriately. Descriptions are present but often generic (avg ~80 chars, below 194-char baseline). Parameters lack depth: no enums for constrained inputs (e.g., git_stash 'action' accepts free-form strings instead of enum ['save','list','pop','apply','drop','clear']), no min/max bounds on numeric params, no output schemas documented. Error handling is minimal, no recovery guidance or actionable error messages visible in code. No tool annotations (readOnlyHint/destructiveHint) despite clear risk stratification (READ_ONLY, WRITE, DESTRUCTIVE). Security concern: tools accept repo_path and repo_url as parameters without visible sanitization against path traversal or command injection.
Create an archive of the repository.
Manage git attributes for the repository.
Show blame information for a file.
Compare two branches and show files changed between them.
Create and/or checkout a branch.
Perform Git clean operations to remove untracked files.
Create a commit with the specified message.
No tool annotations (readOnlyHint/destructiveHint/idempotentHint) despite clear risk stratification. Tools marked DESTRUCTIVE (git_delete_branch, git_reset, git_clean) should declare destructiveHint=true to warn agents. Tools marked READ_ONLY should declare readOnlyHint=true.
Constrained parameters use free-form strings instead of enums. git_stash 'action' should be enum ['save','list','pop','apply','drop','clear']. git_reset 'mode' should be enum ['soft','mixed','hard']. git_config 'scope' should be enum ['local','global','system']. git_archive 'format' should be enum ['tar','zip','tar.gz']. LLMs hallucinate invalid values without enums.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 53 | 2026-07-28+ | v2 |
Get commit history for a branch with optional filtering.
Get detailed information about commits including full messages and diffs.
Configure git settings for the repository.
Create a tag in the repository.
Delete a branch from the repository.
Clone a Git repository and return its directory structure in a tree format.
Manage Git hooks in the repository.
Manage Git LFS (Large File Storage) in the repository.
Fetch Git LFS objects from remote.
Get information about local changes in a repository.
Merge a source branch into the current or target branch.
Pull changes from a remote repository.
Push commits to a remote repository.
Read the contents of specified files in a given git repository.
Rebase the current branch onto another branch or commit.
Reset repository to specified commit or state.
Revert a commit.
Search for code patterns in a repository using git grep.
Stash changes in the repository.
Track (stage) specific files or all files.
No output schemas documented. Tools return results but LLMs cannot plan downstream calls without knowing response structure. E.g., git_commit_history returns commits but response fields (hash, author, date, message) are not specified. Agents cannot extract commit IDs for follow-up operations.
Descriptions are generic and lack actionable context. E.g., 'Get information about local changes in a repository' (git_local_changes) does not explain what 'local changes' means (staged vs unstaged vs untracked). 'Manage git attributes' (git_attributes) is vague, when should an agent call this vs git_config? Descriptions should answer WHAT, WHEN, and WHAT IT RETURNS.
No visible input validation or error recovery guidance. Tools accept repo_path and repo_url as strings without sanitization. No error messages guide LLMs on retry strategy (retryable vs user-fixable vs fatal). E.g., if git_push fails due to auth, the agent should know to ask for credentials, not retry blindly.