VibeDrift plugin
A Claude Code plugin and CLI that keep agent code matching the patterns of the repository and offer MCP checking tools.
- 60 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- Drift Sessions hooks record agent activity once a repository is enabled and `--deep` sends code snippets to the cloud.
- Signed out, anonymous statistics are sent after every scan.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add plugins/vibedriftDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Drift Sessions hooks record agent activity once a repository is enabled and `--deep` sends code snippets to the cloud.
- Signed out, anonymous statistics are sent after every scan.
This is third-party code. Review the repository files before installing.
What it does
VibeDrift looks for drift: gaps between new agent code and the way the repository is already written. Analysis runs locally for JavaScript and TypeScript plus Python, Go and Rust. The `npx @vibedrift/cli` command scans the current folder and opens an interactive HTML report. The MCP server gives the agent eight tools: `get_intent_hints`, `get_dominant_pattern`, `check_file_drift`, `find_similar_function`, `validate_change`, `init`, `enable` and `respond_to_flag`. The agent asks the codebase before writing so new code matches it the first time and finds a similar function instead of writing a duplicate. The Claude Code plugin wires in the server, the drift check skill and the `/vibedrift:setup` command at once.
Who it is for. Teams where an agent writes plenty of code and the repository sprawls into mixed styles and duplicated logic.
Good fit when
- Before creating a function the agent should check whether the project already has a near duplicate.
- You need to learn the dominant repository pattern for a dimension and take an example file.
- A change is proposed and should be checked for drift before it is written.
- You need a quick report over the whole repository with no signup or install.
Not a fit when
- The project language is off the list: JavaScript, TypeScript, Python, Go and Rust are supported.
- You need fully closed work with no network: the cloud deep check requires sign-in.
- You need Drift Sessions on the free tier: they are fully part of the Pro tier.
Example request
Check whether the project already has a function like the one I am about to write and validate the change against the dominant pattern.Limitations
The local engine and local MCP tools are free and need no sign-in. Drift Sessions, `watch`, the pre-push gate and more deep scans stay paid. Signed out, an anonymous beacon leaves after each scan: language, file and line counts, scan time, CLI version, finding count and score, and `--local-only` skips all network calls. The `--deep` mode sends function snippets and relative paths but never whole files. Session hooks capture nothing until you activate a repository with `/vibedrift:setup` or `vibedrift enable`. Node.js 20 or newer is required.
How to disable. Remove the plugin through `/plugin` in Claude Code or run `claude mcp remove vibedrift`. For the CLI run `npm uninstall -g @vibedrift/cli` and delete settings from `~/.vibedrift/`.
FAQ
How do I install the plugin for Claude Code?
In a session run `/plugin marketplace add VibeDrift/VibeDrift` and `/plugin install vibedrift@vibedrift`. The plugin wires the MCP server itself and `claude mcp add` is not needed.
How do I connect the MCP server without the plugin?
Run `claude mcp add vibedrift -- npx -y @vibedrift/cli mcp`. Other MCP clients use the same stdio command: `npx` with arguments `-y @vibedrift/cli mcp`.
What does a `no_baseline` reply mean?
There was no code to analyze or building the baseline failed. The baseline builds on the first tool call in a repository and is then cached.
What goes over the network?
In local mode nothing beyond a daily version check. Signed out, an anonymous beacon follows each scan and `--deep` sends function snippets. `--local-only` turns it all off.
What if `claude mcp add vibedrift` was already run?
After installing the plugin run `claude mcp remove vibedrift` so the server is not wired in twice.
Discussion
No questions or reviews yet.
Related
PluginsA local CLI and read-only MCP server that stop agents from breaking API route layering in Next.js
A Claude Code plugin that hands a task to a local no_human and reviews a branch with a reviewer that never saw the coder's session
A plugin with a local MCP server: the agent scores its own code on a set of criteria through Jev and improves it by the numbers
A Claude Code plugin that generates and maintains CLAUDE.md: a 0 to 100 quality score, a 150-line cap and stale reference checks
CodeDB
codedb
An MCP server and toolset for code in Zig: tree, overview, symbols, search, reading, editing and dependencies. The installer registers it in clients itself
VibeGit is a local version safety net that saves checkpoints before agent edits and rolls a project back after a failure.
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| VibeDrift (this card) | no | 60 | ||
| code-review-graph | no | not required | 32,733 | a local code structure graph serving relevant context over MCP; VibeDrift checks new code against dominant patterns and duplicates |
| AgentLint | no | 156 | a local guard blocking secrets and risky commands; VibeDrift watches code consistency and duplicates |
README in short
The project combines three channels: a CLI a human runs, an MCP server the agent asks and drift sessions with hooks. Analysis is local and the cloud is needed only for deep checks and history.