Memungkinkan AI client untuk memvalidasi dan mengompilasi file spec serta melakukan query pada repository spec dan code graphs melalui local stdio MCP server.
Juga dikenal sebagai specdog/dotdog
- Stars
- 7
- Tools
- —
- Unduhan, 30 hari
- 1.328
- Versi terbaru
- —
- Commit terakhir
- 13 Nov 2026
- Lisensi
- MIT
Apa itu dotdog MCP?
Memungkinkan AI client untuk memvalidasi dan mengompilasi file spec serta melakukan query pada repository spec dan code graphs melalui local stdio MCP server.
Ringkasan dibuat dari README proyek
Tentang dotdog MCP server
Developer yang bekerja dengan coding agents dapat menggunakan dotdog untuk mengelola spec dan code graphs secara local-first. Dengan menjalankan local stdio MCP server melalui command serve, AI client dapat mengakses metadata repository yang terstruktur untuk memahami hubungan antara intensi manusia dan implementasi aktual. Tool ini dirancang bagi mereka yang perlu menjaga lapisan semantik, deklarasi kode yang terobservasi, dan bukti analisis yang diimpor agar tetap terpisah dan dapat di-query dalam workflow pengembangan mereka.
Server ini menyediakan rangkaian tools termasuk search, getEntity, context, impact, dan traverse untuk membantu agent menavigasi codebase yang kompleks. Cara kerjanya adalah dengan memetakan file repository dan mengekstrak deklarasi JS/TS serta static relative imports untuk membangun graph yang disimpan dalam direktori .doghouse. Anda dapat menginstalnya secara global via npm atau Bun, atau menggunakan Homebrew. Tool ini memerlukan Node.js 22.12+ atau Bun 1.3+ dan berfungsi sebagai integrasi berbasis scope repository yang dapat dihubungkan ke Claude Code atau Codex melalui instalasi skill tertentu.
Apa yang bisa Anda lakukan dengannya
- 1Analisis dampak perubahan kodeMinta AI agent untuk menentukan bagaimana modifikasi pada file tertentu akan memengaruhi bagian lain dari sistem.
- 2Query struktur repositoryMinta AI untuk menemukan logika autentikasi atau pola tertentu di dalam codebase yang sudah ada.
- 3Validasi file spesifikasiGunakan agent untuk memeriksa apakah file spesifikasi saat ini sudah benar secara sintaksis dan menyertakan file yang diperlukan.
- 4Impor bukti analisis eksternalBerikan AI client native call graph dari analisis biner untuk memperkaya local code graph.
Ditulis berdasarkan README proyek. Setiap use case diperiksa terhadap kutipan dari README tersebut.
Cara menginstal dotdog MCP
Format diverifikasi 10 Nov 2026Tambahkan ke CursorTambahkan ke GooseInstal dotdog MCP di Claude Code
$ claude mcp add dotdog -- npx -y dotdogInstal dotdog MCP di Cursor
~/.cursor/mcp.json
{
"mcpServers": {
"dotdog": {
"command": "npx",
"args": [
"-y",
"dotdog"
]
}
}
}Instal dotdog MCP di VS Code
.vscode/mcp.json
{
"servers": {
"dotdog": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"dotdog"
]
}
}
}Instal dotdog MCP di VS Code CLI
$ code --add-mcp '{"name":"dotdog","type":"stdio","command":"npx","args":["-y","dotdog"]}'Instal dotdog MCP di Claude Desktop
claude_desktop_config.json
{
"mcpServers": {
"dotdog": {
"command": "npx",
"args": [
"-y",
"dotdog"
]
}
}
}Instal dotdog MCP di Codex
$ codex mcp add dotdog -- npx -y dotdogInstal dotdog MCP di Codex config.toml
~/.codex/config.toml
[mcp_servers.dotdog] command = "npx" args = ["-y", "dotdog"]
Instal dotdog MCP di Gemini CLI
$ gemini mcp add -s user dotdog npx -- -y dotdogInstal dotdog MCP di Gemini settings.json
~/.gemini/settings.json
{
"mcpServers": {
"dotdog": {
"command": "npx",
"args": [
"-y",
"dotdog"
]
}
}
}Pengecekan otomatis
Berjalan 12 Nov 2026Laporan lengkap →Lolos checks otomatis kami. Ini bukan audit keamanan: baca kodenya dan berikan server akses paling terbatas yang dibutuhkannya. Cara kerja checks
- Source repository terhubung dan dapat dijangkauLolos
- Lisensi open source (OSI)Lolos
- Ada commit atau release dalam satu tahun terakhirLolos
- Nama package tidak mirip dengan package populerLolos
- Remote endpoint menggunakan HTTPSTidak diperiksa
- Tools yang bersifat destruktif telah diberi labelTidak diperiksa
- Tidak ada instruksi tersembunyi untuk modelLolos
- Tidak ada advisory kritis pada paket atau dependensinyaTidak diperiksa
README
Lihat di GitHubdotdog
Local-first spec and code graphs for coding agents. Keep human intent, observed implementation, and imported analysis evidence distinct and queryable.
Install
These commands install the stable package. This guide covers Dotdog 1.0.
npm install -g dotdog
# or
bun add -g dotdog
Node.js 22.12+ or Bun 1.3+. npm and Bun install the same package. The global executable has a Node shebang; on a Bun-only machine, use bunx --bun dotdog to force the Bun runtime. The existing Homebrew distribution remains available through the Specdog tap; its formula/version is maintained separately from this source change.
brew tap specdog/tap
brew install dotdog
Existing repository: one refresh, targeted questions
cd your-repository
dotdog refresh --json
dotdog query "authentication"
dotdog context "file:src/auth.ts" --json
dotdog impact "file:src/auth.ts" --json
dotdog path "file:src/auth.ts" "file:src/session.ts" --json
Use an actual graph label from query, not the illustrative paths above. refresh maps repository files, extracts JS/TS declarations and static relative imports, preserves semantic/overlay layers, compiles specs when present, and reports added/changed/removed identities. --if-stale checks content hashes and skips extraction when inputs and output match. Unchanged JS/TS extraction is cached.
Generated repository graphs live under .doghouse/. Do not commit generated repository metadata by default. Human-owned semantic and overlay inputs remain separate from generated facts.
Codex and Claude skills
dotdog skills install --agent both
# or: --agent codex / --agent claude
This installs the bundled dotdog skill into .agents/skills/dotdog/ for Codex and .claude/skills/dotdog/ for Claude Code. It is repository-scoped, idempotent, and refuses to replace user-edited skills. Invoke $dotdog in Codex or /dotdog in Claude; hosts can also select the skill when relevant. Installation does not modify global client settings or install hooks.
Optional automation
| Layer | Responsibility |
|---|---|
| Skill | Choose useful graph context and verify cited evidence. |
| Claude edit hook | Refresh local graph artifacts after supported agent edits. |
| Git hook | Check a temporary snapshot of the staged Git index. |
| CI | Run a pinned version independently of editor behavior. |
| CLI/MCP | Stable execution and query interfaces beneath integrations. |
dotdog hooks install --agent claude
dotdog hooks install --agent git
dotdog hooks status
dotdog hooks remove --agent claude
dotdog hooks remove --agent git
Hooks are explicit opt-in, visible and removable. Existing Git hooks and core.hooksPath configurations are not overwritten. Git hooks never auto-stage files. Claude hooks preserve unrelated settings; edits made through shell commands or external tools may not trigger them. Use refresh --if-stale at the next task boundary and CI as independent checks.
For CI, install a pinned release or build this branch, then run:
dotdog check --json
check rebuilds graphs and checks structural integrity. It does not run application tests or prove runtime behavior. check --staged checks index contents rather than unstaged changes.
New spec project
dotdog init my-project
# Fill in the generated human-owned specs.
dotdog validate
dotdog compile
dotdog design --json
dotdog serve
init refuses an existing project. --minimal creates two files and is supported by validation. Validation reports required-file presence and parser diagnostics, not a percentage of product correctness. Use analyze for heuristic suggestions; use real tests for behavior.
The compact compiler preserves property definitions, states and lifecycle metadata. Incompatible same-named declarations in different source files receive Name@source-file labels; relationships must resolve in their source scope. --v2 retains legacy positional output. Byte-based token figures are estimates, not tokenizer measurements or quality guarantees.
MCP
Configure a local stdio server with the installed dotdog executable and arguments ["serve"], from the intended repository root. It opens no TCP listener.
Tools: listProjects, workspace.list, search, getEntity, schema, summary, traverse, path, context, impact, and infraVerify.
Query an explicit project when more than one is loaded. Observed repository graphs use <repository>:observed names in MCP, keeping them separate from same-named spec projects. Every tool returns structured JSON plus text fallback. Search supports limit and cursor; traversal/impact/context are bounded and disclose truncation. Graph queries retain supplied coverage/limitations; MCP traversal also retains node identities and provenance. no_path means no path in the supplied graph within maxHops, not proof of runtime independence; truncated reports an unexplored search boundary. Invalid graphs are reported separately by listProjects; deleted graphs are evicted.
The server supports legacy initialization revisions 2024-11-05, 2025-06-18 and 2025-11-25, plus 2026-07-28 discovery and per-request metadata. infraVerify is an explicit live network/provider operation; other graph queries stay local.
Selected type contracts
Dotdog 1.0 supports opt-in types inspect for selected TypeScript 7.0.2 contracts, compiler diagnostics and bounded type-reference graphs. See type contracts for commands, requirements and limits. No compiler is auto-installed, and no MCP query executes one.
REA / Ghidra evidence
Dotdog can import a saved REA NativeCallGraph body:
dotdog graph import native-calls.json --format rea-callgraph --project native --artifact-sha256 <binary-sha256> --rea-direction forward
dotdog context "0x1000" --dag .doghouse/imported/native.dag --json
dotdog graph export --dag .doghouse/imported/native.dag --format json
Replace the digest and address with actual evidence values. Supply --rea-direction forward or backward to match the original REA request; the bare export does not encode it. Backward exports are imported as caller → callee edges, not reversed calls. The importer retains traversal coverage, limitations, evidence references, addresses and the export digest. Without a binary digest, identifiers are scoped to the export bytes and binary identity remains unknown. Imported projects are also served by MCP. Imports create separate named graphs; automatic source-to-binary reconciliation and cross-project links are not provided.
This is a saved-export adapter, not a direct Ghidra analysis provider. REA/SolidState retains ownership of Ghidra, target selection, analysis sessions and authority. Dotdog never launches targets, installs REA/Ghidra, or registers another provider. NativeCallGraph bodies are supported; arbitrary REA Evidence envelopes and other Ghidra export formats are not implicitly accepted.
Existing integrations
GitHub Spec Kit remains supported:
dotdog speckit import .
dotdog compile .doghouse/speckit
dotdog serve .doghouse/speckit
Workspaces retain workspace init, add, list, validate, graph, observe, ask, and drift. refresh operates on the current repository; it does not infer cross-repository calls. Use the workspace commands for explicit multi-repo manifests.
Freshness, references and recorded evidence
dotdog status --json
dotdog scenario-check checkout --project shop --json
dotdog reference-check --project shop --json
dotdog graph audit --dag .doghouse/imported/system.dag --profile evidence --json
status is read-only and exits nonzero when graphs are stale or absent. scenario-check checks one named Scenario/Flow for lexical compiled-entity references. reference-check checks explicit Verify file/property mappings without writing human plans. None proves application correctness. staleness, simulate and verify are deprecated compatibility names; verify --init is retired.
graph audit checks recorded artifact hashes, timestamps and results. This establishes artifact identity and reported traceability, not requirement satisfaction or runtime correctness.
Limits and migration
- AST extraction covers JS/TS top-level declarations and static relative imports/re-exports. It does not resolve arbitrary aliases, dynamic calls, runtime execution, or every language's semantics. Other files are inventoried.
- Git ignore rules are honored in Git repositories; non-Git directories use built-in exclusions. There is no silent 500-file cutoff.
impactfollows recorded relationships; it is not a complete runtime blast-radius proof.- 1.0 uses content-derived, case-sensitive identities. Rebuild generated graphs with
refresh; old graph formats remain readable. Do not persist old generated IDs as external contracts without migration. - Missing paths and high-severity drift return nonzero exits, including JSON output.
- No fixed token-savings percentage or hallucination guarantee is claimed.
See retrieval measurements and superposition assessment for the controlled benchmark, exact tokenizer baseline and remaining evaluation limits.
See security review, migration, and integrations.
Development
bun install --frozen-lockfile
bun run lint
bun test
bun run build
node packages/dotdog/dist/cli.js --help
bun run pack:check
bun run release:check
Run the built CLI with node packages/dotdog/dist/cli.js from this checkout. When working in another repository, use the absolute path to that built executable; an existing global dotdog installation may still be the stable release.
See path-search measurements for the tested optimization scope.
The release check packs the actual npm artifact and tests it outside the repository with Node and Bun. MIT licensed.
Author
A specdog project by Justin DiClemente.
Pertanyaan tentang dotdog MCP
Bagaimana cara menambahkan dotdog MCP ke Claude Code?
Jalankan claude mcp add dotdog -- npx -y dotdog di terminal Anda. Tambahkan --scope user agar tersedia di semua proyek.
Bagaimana cara menambahkan dotdog MCP ke Cursor?
Buka ~/.cursor/mcp.json dan tambahkan server di bawah mcpServers, menggunakan config Cursor yang ditunjukkan di bagian instalasi di atas.
Bagaimana cara mengonfigurasi dotdog MCP server?
Install executable dotdog dan arahkan client Anda ke local stdio server dengan argumen ["serve"], dijalankan dari root repository. Server ini tidak membuka TCP listener.
Apa saja MCP tools yang disediakan oleh dotdog?
dotdog menyediakan listProjects, workspace.list, search, getEntity, schema, summary, traverse, path, context, impact, dan infraVerify. Setiap tool mengembalikan JSON terstruktur beserta fallback teks.
Apakah dotdog memerlukan API key atau akun?
Tidak ada autentikasi yang disebutkan; tool ini berjalan secara lokal melalui stdio. Hanya infraVerify yang melakukan panggilan jaringan atau provider secara eksplisit.
Bahasa pemrograman apa saja yang bisa diekstrak kodenya oleh dotdog?
Ekstraksi AST mencakup deklarasi top-level JS/TS serta static relative imports dan re-exports; file lainnya hanya didata dalam inventaris. Tool ini tidak melakukan resolusi pada dynamic calls atau runtime execution.
Apakah dotdog bisa mengimpor hasil analisis Ghidra?
dotdog dapat mengimpor ekspor body REA NativeCallGraph yang tersimpan, dengan tetap mempertahankan cakupan, batasan, dan referensi bukti. Ini bukan merupakan provider analisis Ghidra secara langsung.