vibeDazi MCP Server

中文文档 README_CH.md

Team-shared MCP logging with git user/branch/diffs and prompt context.

Install

./install-vibedazi.sh

Run

vibedazi-mcp

VSCode MCP

Create .vscode/mcp.json:

{
  "servers": {
    "vibedazi": {
      "type": "stdio",
      "command": "vibedazi-mcp"
    }
  }
}

Cursor MCP

Create or update ~/.cursor/mcp.json:

{
  "mcpServers": {
    "vibedazi": {
      "type": "stdio",
      "command": "/absolute/path/to/your/.venv-vibedazi/bin/vibedazi-mcp"
    }
  }
}

Tip: running ./install-vibedazi.sh will create the venv, install the package, and write this file for you using the correct absolute path.

Project policy

The installer writes a root-level rule.md with the policy. Load it into your project prompt (or leave it for teammates to read).

AI Squash Merge tool

This MCP exposes a helper to attribute AI work without cluttering history.

• Tool: ai_squash_merge
• Flow: creates ai/<timestamp>-<hint> branch, commits current changes with AI author, switches back, and git merge --squash into your base branch. Optionally deletes the AI branch.

When To Call

• After you’ve finished an AI-assisted change and validated it (tests/build pass).
• Call it right before your final log_round (so log_round remains last).
• Goal: attribute the work to the AI author without interrupting your development flow.

How To Call

• In Cursor/VSCode MCP Tools: run vibedazi.ai_squash_merge with the params below.
• Or from Python while your venv is active.

Parameters (common)

• branch_hint: optional suffix in branch name
• model_name: label like GPT-5, Claude, o3 (no default; pass explicitly to attribute the right model)
• author_name, author_email: optional overrides for AI identity (if omitted, inferred from model_name)
• base_branch: target branch (defaults to current)
• include_untracked: stage new files too (default true)
• paths: restrict staged paths
• co_author_trailer: add Co-authored-by trailer on squash commit (default true)
• delete_ai_branch: delete the AI branch after squash (default false)
• set_committer_to_ai: set committer to AI on the AI branch commit (default false)
• auto_stash: try stashing if branch creation would overwrite files (default false)

Recommended Usage

• Keep your development on main (or your feature branch). Let the tool branch and return automatically.
• Pass paths to commit only source files (avoid .vibe-dazi/**, __pycache__, *.egg-info).
• Set delete_ai_branch=true to avoid branch clutter unless you want to inspect the AI-only commit later.

Expected Result

• Seamless flow: the tool creates a temporary ai/<timestamp>-<hint> branch, makes a single AI-authored commit, switches back to your base branch, and merges it with --squash to keep history clean.
• Clear attribution:
  • The AI branch commit uses the AI as author (e.g. GPT-5 <noreply@openai.com>).
  • The squash commit includes Co-authored-by and Squashed-from trailers to preserve provenance.
• Uninterrupted development: you remain on your original branch the whole time.
• Auto-cleanup: temporary AI branches are deleted by default (delete_ai_branch=true).

Notes

• If the repository has no commits yet, the tool creates an AI commit on the AI branch and skips the squash (no base to merge into). Make an initial commit and run again to squash.
• On conflict during squash, it returns status: conflict so you can resolve and commit manually.

Examples

• Minimal (auto-detect base branch):
  • ai_squash_merge(branch_hint="refactor", model_name="Claude")
• Commit only code paths and delete the AI branch after:
  • ai_squash_merge(branch_hint="bugfix-xyz", paths=["vibeDazi/src/..."], delete_ai_branch=True)
• If branch creation complains about overwriting files:
  • ai_squash_merge(branch_hint="feature-a", auto_stash=True)

Configuration

You can control defaults and logging behavior via .vibe-dazi/config.json at the repo root (created by the installer). Example:

{
  "log": {
    "ignore_paths": [
      ".vibe-dazi/rounds/**",
      "**/__pycache__/**",
      "**/*.egg-info/**"
    ],
    "max_events": 200,
    "per_file_diff_max_bytes": 120000,
    "snapshot_max_bytes": 240000,
    "snapshot_enabled": true,
    "include_commits_since_last": true
  },
  "ai_squash_merge": {
    "model_name": null,
    "author_name": null,
    "author_email": null,
    "base_branch": null,
    "include_untracked": true,
    "co_author_trailer": true,
    "delete_ai_branch": true,
    "set_committer_to_ai": false,
    "auto_stash": false
  }
}

Notes:

• Change ai_squash_merge.delete_ai_branch to false if you want to keep AI branches around.
• Add patterns to log.ignore_paths to exclude more files from diffs/snapshots.
• per_file_diff_max_bytes and snapshot_max_bytes cap the diff sizes to keep JSON outputs manageable.