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)
Local code validation with explicit execution trust and evidence of what actually ran.
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。
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": {
"checktrail": {
"command": "npx",
"args": [
"-y",
"@stsepelin/checktrail",
"serve",
"--root",
"{project_root}"
]
}
}
}ターミナルでこれを実行します
claude mcp add checktrail -- npx -y @stsepelin/checktrail serve --root '{project_root}'プロジェクトの .cursor/mcp.json、または ~/.cursor/mcp.json に貼り付けます
{
"mcpServers": {
"checktrail": {
"command": "npx",
"args": [
"-y",
"@stsepelin/checktrail",
"serve",
"--root",
"{project_root}"
]
}
}
}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": {
"checktrail": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"@stsepelin/checktrail",
"serve",
"--root",
"{project_root}"
]
}
}
}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。
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7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 stsepelin 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(0.1.0-alpha.5)がレジストリに届いたのは2 か月前です。
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Node.jsも必要です。
同じカテゴリには:Spec Driven Development, Remote Desktop Commander, Systemdox
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.
Your architecture in your agents' context: checks, specs, ADRs and docs over MCP, at write time.
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。
Local code validation with a CLI, MCP tools, and evidence of what actually ran.
The local MCP server works through the chosen MCP host. That host handles AI selection and subscription/API authentication; Checktrail does not need to log in to a model service. Direct provider calls are an optional separate interface. See the MCP review workflow for implemented tools, planned review stages and the limits of fresh-session enforcement.
Formerly Repo Verifier. See the rename guide for existing source checkouts.
Published preview: 0.1.0-alpha.5. Public source is available at stsepelin/checktrail. See implementation status, the plan and the language matrix before relying on an adapter. Installation covers the CLI, Claude Code and Codex. Agent skills provides setup, validation and review workflows installable with npx skills add stsepelin/checktrail. Project setup covers init, doctor and MCP configuration generator. Release preparation records publication and verification gates. The milestone audit separates implemented profiles from open acceptance work; client checks record actual application coverage. Public adoption records the published package on five pinned libraries, including setup friction and compatibility gaps. Alpha.3 fixes the recorded TypeScript 4.9.5 incompatibility. The setup scope guide explains language, documentation and workflow coverage. Alpha.5 adds named Go build-tag profiles with per-check exclusions. Its release record includes package, client and upgrade/rollback verification.
The hosted matrix passed at release commit 832a044 on Linux and macOS. The exact npm tarball and fresh CLI/MCP installations were verified. The MCP Registry entry is active; execution remains disabled by default.
Checktrail discovers projects, plans registered checks, invokes native tools when explicitly trusted, and reports results without turning skipped or empty checks into success. The same engine serves developers, CI and MCP clients.
The optional library task store retains local validation results across restarts. A library worker adds native execution, polling and cancellation. MCP reports still use memory; standard Tasks integration remains pending.
Requires Node.js 22 or newer. Execution currently targets macOS and Linux.
npm ci --ignore-scripts
npm run build node dist/src/cli.js plan --root examples/javascript --detailed node dist/src/cli.js run --root examples/javascript --trust-project --detailed
plan and inspect read files without executing project code. run requires --trust-project: tests, compiler plugins and project configuration can execute code with your user privileges. This is not a sandbox.
Successful commands return JSON except help/version; input errors use stderr. Exit codes:
| Code | Meaning |
|---|---|
| 0 | Selected checks passed, read-only operation succeeded, or advisory experiment completed |
| 1 | At least one selected check failed |
| 2 | Required evidence is incomplete, execution is untrusted, or input is invalid |
A passing run applies to the selected checks and scope only. PHP syntax passing does not imply its application tests passed. Unknown frameworks and empty plans are incomplete. A source change during validation prevents an aggregate pass.
| Ecosystem | Execution in this foundation |
|---|---|
| JavaScript | Native Node tests; explicit ESLint checking with per-file coverage evidence |
| Python | Explicit unittest/pytest tests and Ruff/mypy checks |
| Go | gofmt check, go vet, uncached go test with JSON events |
| PHP | Syntax checks and explicit PHPStan analysis; native verification in status |
| TypeScript | Explicit javascript.typescript: local tsc, no emit, file inclusion evidence |
| Jest | Explicit javascript.jest: native result accounting; pending tests are incomplete |
| Vue | Explicit javascript.vue-tsc: SFC and TS checking; opt-in javascript.vue-router route contracts |
| Vitest | Explicit javascript.vitest: native JSON counts and exact test-file accounting |
| Playwright | Explicit javascript.playwright: native test/project evidence and prepared browsers |
| Other framework test runners | Manifest discovery; execution integrations planned |
| Rust | Locked offline Cargo builds, explicit workspace/target profiles, Clippy/libtest/doctests and Cargo/Rustfmt |
| Ruby | MRI grammar checks; bounded opt-in RuboCop/RSpec/Minitest with source and lifecycle evidence |
| Swift | Grammar checks; bounded Linux ARM64 SwiftPM builds, XCTest/Swift Testing and explicit SwiftLint rules |
| C / C++ | Prepared Clang checks; bounded C17/C++20 CMake builds, CTest, clang-format and clang-tidy |
| Java | Explicit classpath compilation and bounded pinned Checkstyle, Maven reactor and Gradle module tests |
| C# / .NET | Explicit C# compilation; bounded opt-in C#/F#/VB build/NUnit and C#/VB whitespace checks |
| GitHub Actions | Static workflow checking with native per-file accounting |
| Kubernetes | Bounded offline kubeconform document/schema validation with physical source addresses |
| Kustomize | Bounded local resource/prefix/replica assembly with native render reconciliation and source-bound schema findings |
| Terraform | Bounded provider-free multi-file JSON module validation with native source diagnostics |
| Helm | Root-chart and opt-in local application/subchart alias validation with native constraints and source diagnostics |
| Kotlin | Opt-in pinned detekt light analysis with every-selected-file rule, source, suppression and completion accounting |
| Scala | Opt-in pinned Scala 2/3 compilation; explicit mixed/generated/compile-only script and ordered macro stages |
Tools must already be installed. No dependency installation, automatic fixes, service startup, migrations, commits or deployments are performed by the engine. Go execution disables module proxy downloads and toolchain auto-downloads. The Rust profile uses locked offline Cargo and disables rustup auto-installation. Other invoked tools and test code can still access the network.
Without configuration, registered defaults apply to each detected project. Node tests are selected automatically only for the exact script node --test. Python needs an explicit selection because a manifest does not identify a runner. Create checktrail.json in the inspected root:
{
"schemaVersion": 1, "projects": [ { "path": "web", "checks": ["javascript.node-test", "javascript.eslint"] }, { "path": "library", "checks": ["javascript.typescript"] }, { "path": "service", "checks": ["go.format", "go.vet", "go.test"] }, { "path": "worker", "checks": ["python.unittest"] } ] }
When present, this file selects only the listed projects/checks. Paths are exact discovered project roots relative to --root; use . for that root itself. Configuration cannot supply arbitrary commands or enable execution permissions. The schema rejects unknown fields and check IDs are validated by the engine. Private configuration and rules need not be published. Local public/private JSON packs and additive operator overlays are implemented; see policy packs. An explicit fetch-pack command can download a pinned data-only pack over HTTPS for later offline use. The explicit fetch-adapter command verifies every artifact in a pinned executable bundle before publishing a private file; registration and trusted execution are separate actions.
Operator-registered external adapters can run a pinned local Node, Python, PHP or native executable bundle. Repository configuration can select their registered checks; execution still requires operator trust. The protocol checks evidence completeness, while the adapter author remains responsible for the correctness of its analysis.
The TypeScript check requires a local tsconfig.json and an installed compiler inside the configured root (project-local or hoisted to an ancestor). It disables noCheck, emission and incremental state. Every inventoried .ts, .tsx, .mts and .cts file in that project must appear in the compiler's file list. A clean compiler exit with excluded source is incomplete. Other compiler options, including strictness and declaration checking, come from the project config. Use javascript.typescript-build for project-reference solutions with fresh in-memory declarations. Vue SFC checking uses its separate adapter. TypeScript 6.0.3 is exercised in integration tests; other versions are unverified.
javascript.vue-router captures a selected native testing router and verifies complete record participation in declared URL probes; see the Vue Router profile for versions and projection limits. javascript.nuxt-runtime builds a fresh SSR testing assembly and checks native router capture after in-process requests. Opt-in version 2 also verifies native server handlers, middleware invocations, runtime config boundaries, generated API consumer types and controlled responses. See the Nuxt profile and assembly contracts for versions and scope limits. python.fastapi-routes captures a configured FastAPI application's native routes. Opt-in configuration version 2 also checks declared mounted/hosted assembly, security dependencies, response models, middleware/lifespan order and controlled ASGI responses. See the FastAPI profile and assembly contracts for versions and scope limits. python.django-routes captures supported nested URL resolv
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