Spec Driven Development
MCP Server that facilitates spec-driven development workflows, not just Vibe Coding.
作者による説明(英語の原文)
Personalize AI coding assistants by analyzing local conversation history. Runs on your machine.
この3つを1つのスコアにまとめることはしません。意見、測定、人気はそれぞれ別のものだからです。
このサーバーのやさしい解説はまだ書いていません。以下はレジストリから読み取った事実です。
使っているAIアプリを選んで、設定をコピーしてください。
VARDOGER_MCP_PLATFORM必要なもの
どのアプリを使っていますか?
上でアプリを選んでください。選んだアプリを覚えておくので、次からはボタンに「…にインストール」と表示されます。
Claude Desktop → Settings → Developer → Edit Config を開き、claude_desktop_config.json にこれを貼り付けます
{
"mcpServers": {
"vardoger": {
"command": "uvx",
"args": [
"vardoger",
"mcp"
],
"env": {
"VARDOGER_MCP_PLATFORM": "your-value"
}
}
}
}ターミナルでこれを実行します
claude mcp add vardoger --env VARDOGER_MCP_PLATFORM=your-value -- uvx vardoger mcp動かす前に、空欄の値を入力してください。
プロジェクトの .cursor/mcp.json、または ~/.cursor/mcp.json に貼り付けます
{
"mcpServers": {
"vardoger": {
"command": "uvx",
"args": [
"vardoger",
"mcp"
],
"env": {
"VARDOGER_MCP_PLATFORM": "your-value"
}
}
}
}ブラウザがアプリを開いてよいか確認し、そのあとアプリが設定を表示して確認を求めます。キーの値は空欄のままなので、自分で入力してください。
プロジェクトの .vscode/mcp.json に貼り付けます
{
"servers": {
"vardoger": {
"type": "stdio",
"command": "uvx",
"args": [
"vardoger",
"mcp"
],
"env": {
"VARDOGER_MCP_PLATFORM": "your-value"
}
}
}
}ブラウザがアプリを開いてよいか確認し、そのあとアプリが設定を表示して確認を求めます。キーの値は空欄のままなので、自分で入力してください。
これはパソコンで動くサーバーですが、ChatGPTがつなげるのはインターネット上のサーバーだけです。ほかのアプリを選んでください。
選んだアプリは、この端末のCookieに保存されます。アカウントは不要です。
VARDOGER_MCP_PLATFORM envこの設定は、作者のレジストリ登録情報から作成しています。READMEと違う場合は、READMEに従ってください。
パソコンで動くサーバーです。インストールすると、ツール一覧がAIアプリに表示されます。
誰でも見られる情報をチェックしており、コードは読んでいません。当てはまらない項目は数に入れていません。 · 2026年11月6日にチェック
レジストリ名は、リポジトリを所有するGitHubアカウント dstrupl で確認済みです
このリポジトリはまだGitHubから読み取っていません
このリポジトリはまだGitHubから読み取っていません
これはパソコンで動くため、インターネット上のエンドポイントはありません
パソコンで動き、代わりにローカルの設定を使います
パソコンで動くサーバーです。ツール一覧を読み取れるのは、インターネット上のエンドポイントがある場合だけです
レジストリでこのカテゴリにある有効なすべてのサーバーと比較しています。スターは、公開者がリポジトリの所有者である場合だけ数えます。
当てはまる基本チェック1項目のうち1項目に合格しました。コードは読んでいません。キーやトークンを追加する前に、サーバーがどんなアクセスを求めているかを確認し、必要な権限だけを持つキーを使ってください。
レジストリには必須のキーは載っていません。ただし、最初に使うときにサインインを求めるサーバーもあります。
あなたのパソコンで動きます。必要なときにAIアプリが起動し、データは作者のサーバーを通りません(サーバー自体がオンラインサービスを呼び出す場合を除きます)。
GitHubリポジトリの最終更新は–、最新バージョン(0.4.0)がレジストリに届いたのは40 日前です。
Claude Desktop、Claude Code、Cursor、VS CodeなどのMCP対応アプリで使えます。パソコンにPython (uv)も必要です。
同じカテゴリには:Spec Driven Development, Remote Desktop Commander, Recta
同じカテゴリで、ほかの作者が作ったサーバー。スターの多い順。
カードの見方:★ GitHubスターi3つの柱:レビュー · チェック · 利用状況i更新=最新バージョンをレジストリに公開した日i
MCP Server that facilitates spec-driven development workflows, not just Vibe Coding.
Remote Desktop Commander: use files, terminal and processes on your own computers from any AI app.
Cheaper cross-chain stablecoin routes for AI agents; 5 % of the verified saving, nothing if none.
作者のリポジトリから書かれたとおりにコピーしたものです。PopNx MCPは内容を確認・推奨していません。HTML、画像、スクリプトは取り除いています。リンクはこのサイトの外へ移動します。
A cross-platform plugin for AI coding assistants (Cursor, Claude Code, OpenAI Codex, OpenClaw, GitHub Copilot CLI, Devin Local, legacy Windsurf, and Cline) that reads supported local conversation-history formats, extracts behavioral patterns, and generates personalized system prompt additions — making the assistant progressively better suited to how you work. Current format limitations are called out per platform below rather than silently reading undocumented stores.
History discovery, parsing, checkpoints, and generated rules stay on your machine. Analysis is performed by the host assistant you invoke, so selected conversation excerpts follow that assistant and model provider's data policy. Vardoger operates no backend and sends no telemetry.
| Platform | Command |
|---|---|
| macOS | brew install python@3.13 (install Homebrew) or python.org/downloads/macos |
| Debian / Ubuntu | sudo apt install python3 |
| Fedora | sudo dnf install python3 |
| Windows | winget install Python.Python.3.13 or python.org/downloads/windows |
Recommended for installing vardoger as an isolated CLI tool. Full instructions at pipx.pypa.io/stable/installation.
| Platform | Command |
|---|---|
| macOS | brew install pipx && pipx ensurepath |
| Debian / Ubuntu | sudo apt install pipx && pipx ensurepath |
| Fedora | sudo dnf install pipx && pipx ensurepath |
| Windows | scoop install pipx or pip install --user pipx && pipx ensurepath |
pipx install vardoger
vardoger setup cursor # or claude-code, codex, openclaw, copilot, devin, windsurf, cline
Then tell your assistant: "Personalize my assistant."
Looking for the in-app plugin listings? Track review status for each
marketplace (PyPI, Cursor, Claude Code, Codex, Copilot CLI, Devin Desktop,
Cline,
ClawHub) in
MARKETPLACE_STATUS.md.
Vardoger is currently live on PyPI, the Claude Code community catalog,
self-hosted Codex and Copilot marketplaces, McpMux, and the Official MCP
Registry. The former Cursor listing is currently unavailable. ClawHub has
regressed to 0.3.1 and is intentionally frozen because its mandatory MIT-0
terms conflict with this Apache-2.0 project. Current OpenClaw history has a
live-accepted, opt-in full-read Gateway CLI path through the direct install.
Previous pre-releases.
pipx install vardogernow resolves to the stable
0.4.0release. The beta install paths below stay here for anyone still pinning
an earlier release; new installs should not need them.
`bash
# opt into future pre-releases (0.2.0bN, etc.):
pipx install --pip-args="--pre" vardoger
# or pin an older pre-release:
pipx install vardoger==0.1.0b3
# or run without installing:
uvx vardoger --help
`
| Command | Purpose | |
|---|---|---|
vardoger setup | Register vardoger with a platform (cursor, claude-code, codex, openclaw, copilot, devin, windsurf, cline). | |
vardoger status [--platform X] [--json] | Report whether each personalization is fresh or stale. | |
vardoger prepare --platform X [--batch N] [--synthesize] | Produce the batched prompts used by the AI-driven skill pipeline. | |
vardoger write --platform X | Read synthesized personalization from stdin and write it to the platform's rules file (supports YAML-frontmatter confidence metadata). | |
| `vardoger feedback accept\ | reject --platform X [--reason TEXT]` | Record whether you kept or rejected the last generation. reject auto-reverts to the prior generation. |
| `vardoger compare --platform X \ | --all [--window DAYS] [--json]` | Compare heuristic conversation-quality metrics before vs. after the latest personalization. |
vardoger profile sources [--platform X] [--json] | List saved generations and their stable one-based selection IDs. | |
vardoger profile preview --source X:N [--source Y:latest] | Compile explicitly selected generations into a normalized cross-host profile and show the portable AGENTS.md diff without writing. | |
vardoger profile write --source X:N --target ./AGENTS.md [--apply] | Show the same diff; update only Vardoger's fenced block when the user explicitly adds --apply. |
.cursor/rules/, .claude/rules/, AGENTS.md, etc.)First run vs. incremental runs. By default vardoger does not apply a
time window — the first run reads your full local history (that is when the
signal is richest and a windowed default would silently drop older sessions
you never get a second chance to feed in). After that, a per-conversation
checkpoint store at
~/.vardoger/state.jsonensures every subsequent run
only reprocesses new or changed conversations, so refreshes stay fast.
If you have very large local history and want to cap the first-run cost,
pass
--since DAYS(e.g.vardoger prepare --platform cursor --since 90);
use
--fullto force a full re-crawl that bypasses the checkpoint.
The profile compiler combines only existing Vardoger generations that you select; it does not silently reopen raw transcripts. Generation indexes are one-based, while latest resolves to the newest saved generation for that platform:
# Discover the reviewed generations available for selection.
vardoger profile sources
vardoger profile preview \ --source claude-code:latest \ --source codex:2 \ --target ./AGENTS.md
vardoger profile preview \ --source codex:latest \ --min-confidence medium \ --max-age 180 \ --redact '@example\.com' \ --target ./AGENTS.md
vardoger profile write \ --source claude-code:latest \ --source codex:2 \ --target ./AGENTS.md \ --apply
The normalized JSON audit is available with profile preview --json. It records source platform, generation, output hash/path, observation time, recency, confidence, conflicts, supersession, and exclusions. Opposing rules such as “Prefer tabs” and “Avoid tabs” are withheld from active instructions until reviewed. Writes preserve everything outside and .
| Platform | History Source | Prompt Delivery | Integration |
|---|---|---|---|
| Cursor | Agent transcript JSONL | .cursor/rules/vardoger.mdc | Marketplace plugin + MCP |
| Claude Code | Session JSONL | .claude/rules/vardoger.md | Community/custom plugin |
| OpenAI Codex | Session rollout JSONL | ~/.codex/AGENTS.md | Repository marketplace plugin |
| OpenClaw | Legacy JSONL or opt-in full read through official Gateway CLI; private SQLite is never queried | ~/.openclaw/skills/vardoger-personalization/SKILL.md | Analyzer + Gateway compatibility status |
| GitHub Copilot CLI | ~/.copilot/session-state/ /events.jsonl plus legacy flat JSONL | ~/.copilot/copilot-instructions.md (global) or /.github/copilot-instructions.md (project) — managed inside a fenced section | Custom plugin + published skill |
| Devin Local | User-enabled Devin CLI --export ATIF JSON under ~/.vardoger/imports/devin/ | ~/.config/devin/AGENTS.md (global, fenced section) or /.devin/rules/vardoger.md | ATIF + skill + rules integration |
| Legacy Windsurf / Cascade | ~/.codeium/windsurf/**/*.jsonl | ~/.codeium/windsurf/memories/global_rules.md (global, fenced section) or /.windsurf/rules/vardoger.md (project, dedicated file) | Legacy Cascade skill + CLI + MCP |
| Cline | VS Code globalStorage/.../tasks/*/api_conversation_history.json | ~/Documents/Cline/Rules/vardoger.md (global) or the existing .clinerules project layouts | CLI + MCP |
Requires uv (Python package manager):
git clone https://github.com/dstrupl/vardoger.git
cd vardoger uv sync .venv/bin/vardoger --help
src/vardoger/ # shared core — history reading, analysis, prompt generation
.agents/plugins/ # repository-level Codex marketplace catalog plugins/_shared/ # shared analysis/personalization skill authored once plugins/cursor/ # Cursor MCP server config, install script plugins/claude-code/ # Claude Code plugin manifest, skills plugins/codex/ # Codex plugin manifest, skills plugins/openclaw/ # OpenClaw skill plugins/copilot/ # GitHub Copilot CLI plugin manifest, skills plugins/devin/ # Devin Local ATIF export, skill, and rules integration plugins/windsurf/ # Legacy Cascade install snippet and rules delivery plugins/cline/ # Cline integration and marketplace install guidance tests/ # all tests, mirroring src/ structure
src/vardoger/.plugins/ /.tests/, mirroring the source tree.See AGENTS.md for full coding standards and quality checks.
CI enforces a combined quality bar on every push and pull request:
ruff check / ruff format --check — lint (incl. complexity, pylint, return, pathlib, tryceratops rules) and formatting.mypy src/ — strict type checking.pytest --cov=vardoger --cov-fail-under=80 — tests across Python 3.11–3.13 with a combined 80% coverage floor.bandit -r src/ and pip-audit --skip-editable to catch common code smells and dependency CVEs.Run the full bundle locally before pushing:
uv run ruff check . && uv run ruff format --check . && uv run mypy src/ && uv run pytest --cov=vardoger --cov-fail-under=80
Contributions are welcome. Short version:
dstrupl/vardoger on GitHub and clone your fork.uv sync and create a topic branch.main.CI (test on Python 3.11/3.12/3.13 plus a security job) will run automatically on the PR. First-time contributors may need a maintainer to click Approve and run before the first workflow execution.
See CONTRIBUTING.md for the full walkthrough and AGENTS.md for coding standards and commit-message conventions.
CI runs automatically on every push and PR (lint, type check, tests across Python 3.11–3.13). To publish a new version:
CHANGELOG.md.main, then wait for all required checks.vX.Y.Z tag and publish the matching [GitHub release](https://github.com/dstru冒頭だけを表示しています。続きはリポジトリで読めます