HoloScript MCP server
2026년 11월 3일 커밋백엔드 서비스, 로봇, 공간 애플리케이션 등 다양한 타겟을 위한 시맨틱 시스템 프로그래밍 코드를 AI agent가 작성하고 컴파일할 수 있도록 지원합니다.
다른 이름 brianonbased-dev/HoloScript
- 스타
- 11
- Tools
- —
- 다운로드, 30일
- —
- 최신 버전
- —
- 마지막 커밋
- 2026년 11월 3일
- 라이선스
- MIT
HoloScript MCP란?
HoloScript은(는) MCP server입니다. 백엔드 서비스, 로봇, 공간 애플리케이션 등 다양한 타겟을 위한 시맨틱 시스템 프로그래밍 코드를 AI agent가 작성하고 컴파일할 수 있도록 지원합니다.
프로젝트 README에서 생성된 요약
HoloScript MCP server 정보
HoloScript는 개발자와 AI agent가 다양한 배포 타겟을 위한 시맨틱 시스템 프로그래밍 코드를 작성할 수 있게 해줍니다. 플랫폼별로 별도의 연결 코드를 작성하는 대신, 사용자는 .holo, .hsplus 또는 .hs 파일을 사용하여 시스템을 한 번만 기술하면 백엔드 서비스, 로봇, 공간 애플리케이션 또는 웹 런타임을 생성할 수 있습니다. 이 언어는 결정론적 컴파일러를 사용하여 소스 코드를 C++, TypeScript, Rust 또는 엔진 전용 코드와 같은 다양한 출력물로 변환하며, 모든 타겟에서 원래의 의도가 유지되도록 보장합니다.
AI agent는 MCP 도구를 통해 이 서버를 사용하여 시스템을 검사하고, 제한된 범위 내에서 변경을 수행하며, 작업 내용을 보고합니다. 설정은 npx를 사용하여 로컬에서 구현하거나 원격 연결을 통해 수행할 수 있습니다. 공개 도구를 사용하면 IP당 분당 30회 요청 제한 내에서 읽기 전용 파싱, 유효성 검사 및 구문 도움을 받을 수 있으며, 사용자는 클라이언트를 등록하여 전체 도구 세트에 액세스할 수 있습니다. 이러한 방식을 통해 단일 소스 파일이 게임 엔진 빌드, 실행 중인 서비스 또는 디지털 트윈과 같이 서로 관련 없는 여러 결과물로 변환될 수 있습니다.
활용 사례
- 1멀티 타겟 공간 애플리케이션 개발개발자는 Unity, Unreal, Godot 또는 WebGPU 타겟으로 컴파일되는 단일 장면을 기술할 수 있습니다.
- 2agent를 통한 자동화된 시스템 오케스트레이션AI agent는 MCP 도구를 사용하여 시스템을 검사하고 검증된 변경을 수행할 수 있습니다.
- 3통합 백엔드 및 서비스 배포팀은 서비스를 한 번 정의하고 이를 다양한 백엔드 또는 서비스 타겟으로 컴파일할 수 있습니다.
- 4로보틱스 및 시뮬레이션 모델링엔지니어는 로보틱스 또는 시뮬레이션 타겟으로 직접 내보낼 수 있는 로봇 구성을 기술할 수 있습니다.
프로젝트의 README를 바탕으로 작성되었습니다. 모든 활용 사례는 해당 내용의 인용문과 대조하여 확인합니다.
자동 검사
2026년 11월 14일에 실행됨전체 보고서 →아직 인증되지 않았습니다. 서버는 80점 이상의 점수, 심각한 결함 없음, 그리고 검사에 충분한 데이터를 갖추어야 합니다. 검사 방식 · 점수를 높이기 위해 수정할 항목
- 소스 저장소 연결 및 접근 가능brianonbased-dev/holoscript is public and reachablegithub.com/brianonbased-dev/holoscript ↗통과
- 패키지가 이 저장소를 소스로 명시검사되지 않음
- 빌드 출처(Sigstore 증명)검사되지 않음
- 오픈 소스 라이선스 (OSI)MITspdx.org/licenses/MIT.html ↗통과
- 최근 1년 내 커밋 또는 릴리스Last commit or release 2026-11-03github.com/brianonbased-dev/holoscript ↗통과
- OpenSSF Scorecard검사되지 않음
- 패키지 이름이 인기 패키지와 유사하지 않음검사되지 않음
- 패키지 및 해당 종속성에 중요한 보안 권고 없음검사되지 않음
- 모델에 대한 숨겨진 지침 없음0 tool descriptions and the README read; no override phrases, credential paths or hidden characters통과
- 파괴적인 도구에 라벨 표시됨검사되지 않음
- 원격 엔드포인트가 MCP 핸드셰이크에 응답검사되지 않음
README
GitHub에서 보기HoloScript
HoloScript is a general-purpose semantic systems programming language: you describe a system once, and real compilers — not a language model — lower that same source into an engine build, a backend service, a robot description, a web runtime, or a headset app.
It programs applications, services, runtimes, simulations, agents, devices, and spatial worlds from source files you own. Declarative composition and traits are language mechanisms, not a domain boundary. The language is under active construction; the systems-language bar below states plainly which parts of that identity are implemented and which are not.
Humans and AI agents write .holo, .hsplus, or .hs programs for screens, scenes, services, data, devices, runtimes, and agent workflows. HoloScript can interpret those programs while they evolve, lower them through its own VM and sovereign backends, or compile bridge artifacts for an external browser, engine, robot, service, or deployment target. The source remains HoloScript; generated C++, TypeScript, Rust, engine code, and interchange formats are outputs rather than the language's ceiling.
Think of it as three practical pieces:
- a programming language whose source preserves semantics, behavior, effects, and constraints
- runtimes and VMs that execute that source without requiring an external engine
- compilers that lower the same program into native, sovereign, and compatibility targets
Commonly gotten wrong
The name misleads, and summaries of this project usually inherit the mistake. HoloScript is not a scripting language. The -Script suffix files it next to JavaScript and GDScript, and it does not belong there.
| Often read as | Actually |
|---|---|
| a scripting language | a general-purpose systems language with its own lexer, parsers, AST, type and trait system, VM, and compiler backends |
| a hologram or 3D tool | spatial output is one compile target among many — .hs pipelines and service compositions have no spatial surface at all |
| a scene format or DSL | a language with effects, constraints, and execution semantics; scenes are one thing it can describe, not the boundary of what it is |
| an AI that writes code for you | agents author HoloScript through MCP tools, but the compilers that turn it into artifacts are deterministic — a model never emits the output |
| a wrapper over Unity or Unreal | engines are bridge targets; the same source also runs on HoloScript's own VM and sovereign runtimes without an external engine |
The distinguishing claim is not that you can write code here. It is that one source file becomes many unrelated kinds of artifact — a game engine build, a running service, a robot, a digital twin, a headset app — through real compilers you can inspect.
Facts at a glance
Verified against this source tree (counts drift upward; verify with the commands in CLAUDE.md Knowledge Pack):
- Three source formats:
.hs(declarative scenes/logic),.hsplus(fully typed programs),.holo(whole-system compositions and UI surfaces) - 80+ compiler backends in
@holoscript/corelowering one source to 25+ documented targets — Unity, Unreal, Godot, VRChat (Udon), WebGPU, Three.js/R3F, Babylon.js, PlayCanvas, WASM, iOS, visionOS, Android XR, OpenXR, robotics (URDF/SDF), IoT (DTDL/WoT), neuromorphic, A2A (docs/compilers/) - 1,500+ semantic trait implementations across 150+ categories (docs/traits/)
- Full toolchain: hand-built lexer/parsers/AST, LSP, formatter, linter, CLI, runtime, in-browser WASM parser, Python bindings
- Deterministic compilation: AI agents author HoloScript through MCP tools; real compilers — not an LLM — produce the output
- Progressive self-hosting: the language's own lexer and parsers have
.hsplusimplementations alongside the TypeScript bootstrap, and holoscript.net is authored in.holoand compiled by HoloScript's own NextJSCompiler
Use this to connect an AI coding agent to HoloScript tools:
{
"mcpServers": {
"holoscript": {
"command": "npx",
"args": ["-y", "mcp-remote", "https://mcp.holoscript.net/mcp"]
}
}
}
Try it without an API key
npx create-holoscript@latest my-app- scaffold a HoloScript app with a browser scene in under a minute.npm install @holoscript/cli- use the local CLI;npm install @holoscript/wasmadds in-browser parsing.- https://holoscript.studio/playground - try the live editor and WebGPU preview with no login.
POST mcp.holoscript.net/api/public/tool- call read-only public tools for parsing, validation, examples, syntax help, explanations, and target lists. 30 req/min per IP.POST mcp.holoscript.net/oauth/register- register a client yourself when you need the full set of tools. No human approval step.
Requires npm 11+ for now. Installing the CLI with npm 9 or 10 currently fails (see the known issue below). To install it without changing your global npm:
npx -y npm@11 install @holoscript/cli@8.9.0
npx -y npm@11 exec -p @holoscript/cli@8.9.0 -- hs --version
The first command installs into the current folder.
Or upgrade npm first with npm i -g npm@11, then run npm install @holoscript/cli@8.9.0 as usual.
Known issue: on npm 9 and 10 (bundled with Node 18 to 22), npm install @holoscript/cli and npx -p @holoscript/cli hs fail with Cannot read properties of null (reading 'edgesOut'). The cause is that @holoscript/absorb-service@6.1.5 lists dev tools (vitest) as runtime dependencies. This is fixed in source by #667; a patched release is pending.
Full reference: docs/PUBLIC_ACCESS.md.
Why build systems in HoloScript
HoloScript helps when the same product idea has to show up in too many places: a browser preview, engine export, service API, agent workflow, robotics simulation, internal tool, or digital twin. Instead of burying the idea in framework-specific glue, it keeps the intent in a readable source file that humans can review, agents can edit, and HoloScript can run.
1) Program semantics, not framework glue
Use .holo, .hsplus, or .hs files to define behavior, data, interfaces, objects, and workflows. Run them directly, then translate them to platform code when that is worth it.
2) Give agents something solid to work with
HoloScript includes task boards, MCP tools, connectors, permissions, and traits so agents can inspect a system, make bounded changes, and report what they did.
3) Keep deployment paths open
The same composition can feed multiple outputs: renderers, services, robotics, web runtimes, and more. If a translation is not ready, HoloScript can still run the source.
4) Compose real product capabilities
Add production concerns as traits instead of one-off glue:
- payments and economy
- compliance and auditability
- search and data pipelines
- collaboration and synchronization
- observability and incident response
- AI/ML and agent orchestration
- IoT, simulation, and digital twins
5) Start with one use case, grow from there
Start with a dashboard, workflow, agent task, or 3D scene. Add traits and plugins as the work grows, without rewriting the core idea.
HoloScript is intended to occupy the full general-purpose systems-language layer. External engines, operating-system APIs, hardware SDKs, and interchange formats remain useful bridge targets, but they do not define the language. New language power belongs in sovereign parsers, runtimes, VMs, standard libraries, and compiler backends.
Systems-language bar
The identity is ahead of some implementation layers. HoloScript does not claim systems-language closure merely because it can generate target code. The remaining bar includes an explicit memory and resource model, stable data layout and ABI/FFI contracts, concurrency and effect semantics, native code generation, debugger/profiler support, and progressive self-hosting. Current status belongs in the language identity contract and spec-versus-reality ledger.
Who this is for
- Product teams: a faster path from idea to running prototype.
- Platform engineers: one readable source for outputs that would otherwise split across stacks.
- Agent builders: files, tools, and checks agents can inspect before acting.
- Research / simulation teams: developer-facing solvers and repeatable workflows with provenance and replay hooks.
One-line outcomes by role
| If you are a... | Describe this | Get this |
|---|---|---|
| Game / XR developer | composition "Dungeon" { object "Chest" @grabbable @physics { ... } } |
Your scene runs on Unity, Unreal, Godot, or React Three Fiber from one source you own — you pick the target, you're never locked to it |
| AI agent builder | agent "Brittney" { tool: generate_scene, tool: deploy_service, ... } |
MCP tools with typed inputs, permissions, and live inventory verified via GET /health |
| Simulation engineer | simulation "WindTunnel" { solver: fea, mesh: tet10, boundary: ... } |
TypeScript-accessible solvers with replay/provenance hooks; verify geometry, meshing, boundary conditions, and V&V depth before scientific use |
| Founder / product team | service "BillingAPI" { route: /invoice, method: POST, ... } |
Node.js service scaffold with observability, metering, and rollback hooks |
Example outcomes
- "Describe a service" -> compile to backend/service targets
- "Describe a scene" -> compile to spatial/render targets
- "Describe a robot" -> export to robotics/simulation targets
- "Describe an agent workflow" -> execute via board + tool ecosystem
Example stories in the repo
The example files are part of the pitch. They show how the three HoloScript formats carry different kinds of intent:
| Format | Example | What it helps someone picture |
|---|---|---|
.hs |
examples/pipelines/inventory-sync.hs |
A scheduled POS inventory sync: source, transform, validate, and sink. |
.hsplus |
examples/three-format-showcase/smart-gallery.hsplus |
A gallery that grows from objects into templates, state, interactions, and audio. |
.holo |
examples/iot/holotwin-smart-farm.holo |
A smart farm dashboard where MQTT sensors drive a 3D digital twin. |
.holo |
examples/novel-use-cases/05-robot-training-metaverse.holo |
Aspirational grammar (examples health matrix; parser fail). Not a working HoloScript example. Story only: a robot training arena that links simulation, agents, feedback, and ROS-style targets. |
| All three | examples/three-surface-agent/ |
One agent whose composition, cognition, and typed policy execute under one receipt. |
Use these when explaining HoloScript to a new person: start with the story, then show the format that carries it.
What to verify live
HoloScript avoids hardcoded ecosystem counts in docs. Verify current numbers from live sources:
- MCP tool count:
https://mcp.holoscript.net/health - Knowledge entries:
https://mcp-orchestrator-production-45f9.up.railway.app/health - Compiler/trait/package counts: see SSOT commands in
docs/NUMBERS.md
Honest gaps
These verticals have foundation traits but no dedicated domain coverage yet:
| Gap | Closest Existing | Distance |
|---|---|---|
| Geolocation / GIS | GeospatialAnchorTrait, RooftopAnchorTrait |
Close — AR anchoring exists, mapping layer doesn't |
| Calendar / Scheduling | @cron, @scheduler, @task_queue |
Medium — job scheduling exists, calendar UI doesn't |
| CRM | @tenant, @session, @analytics |
Far — primitives exist, CRM workflow doesn't |
| Inventory | @database, @data_transform |
Far — data traits exist, inventory domain doesn't |
| Logistics / Shipping | SCMCompiler (supply chain) |
Medium — compiler exists, trait coverage thin |
| Real Estate | Spatial rendering + @digital_twin |
Medium — visualization ready, domain traits missing |
| Agriculture | @iot_sensor, @digital_twin, @telemetry |
Medium — IoT foundation covers hardware |
| Energy / Utilities | @iot_sensor, @mqtt_bridge, @digital_twin |
Medium — same IoT foundation |
| Legal / Contracts | @approval, @audit_log, @consent_management |
Medium — compliance exists, legal workflow doesn't |
| Government / Civic | @audit_log, @rbac, @gdpr |
Medium — compliance traits, no civic domain |
Every gap shares the same pattern: the infrastructure traits exist, the domain plugin doesn't. The plugin system (packages/plugins/) is designed exactly for this — add domain-specific traits without touching core.
Vertical README epics (optional backlog): some packages have deep READMEs, others are stubs. Tracked in docs/vertical-readme-epics.md for prioritization; split into one board child task per vertical when a row is ready to ship.
Architecture — the 6-layer stack
HoloScript is a 6-layer open platform. The same model anchors the landing page and this repo, so the story reads the same whether someone (or an agent) starts from the site or the source.
Universal use has a boundary: builders normally consume HoloScript through project files, npm packages, MCP/API, Studio workspaces, or service images; engine contributors clone this monorepo to change the platform itself. The full contract lives in docs/architecture/universal-use-boundary.md.
| Layer | What it is | Where it lives |
|---|---|---|
| 6 · Marketplace | Traits + plugins + skills, Ed25519 signatures, x402 payments | store.holoscript.net, packages/plugins/ |
| 5 · Studio | AI scene builder, browser workflow, node-graph editor | packages/studio/ |
| 4 · Agents | Swarm intelligence, MCP comms, economy primitives | packages/mcp-server/ |
| 3 · Compiler | Named backends, cross-target compilation, circuit breaker, incremental builds | packages/core/src/compiler/ |
| 2 · Runtime | Three.js browser, Rust spatial engine, WebGPU rendering | packages/r3f-renderer/, packages/snn-webgpu/, packages/compiler-wasm/ |
| 1 · OS Core | Cognitive VM (uAAL, stratum ③), perceptual (SNN), economic (x402), semantic traits | packages/core/ |
The layers are architectural, not counts — they don't go stale. Ecosystem counts (tools, traits, compilers) are never hardcoded here; see What to verify live.
How it composes (the loops)
Individual traits solve individual problems. When you wire them together, autonomous loops emerge:
Description → running business. .holo → generate_service_contract → compile_to_node_service → connector-railway (deploy) → economy (meter) → observability (monitor) → self-improve (iterate). One file produces a deployed, metered, monitored, improvable service.
Self-healing infrastructure. Circuit breaker + health check + canary + rollback + incident + agent orchestration + PagerDuty. Detects its own failures, routes around them, rolls back, recovers — no human in the loop.
Recursive knowledge compounding. Absorb scans code → extracts W/P/G → publishes to knowledge store → other agents consume → write better code → absorb scans again. Every cycle, the ecosystem knows more.
Agent economy. Agent creates trait → publishes to marketplace with signing → another agent discovers, installs, pays → revenue splits → creator reinvests. Agents paying agents for work product.
Cross-reality coordination. CRDT + presence + sync + connectors. Same composition state live in VS Code, browser, phone, headset, and IoT device simultaneously — with conflict resolution.
Full provenance chain. CAEL (Continuous Agentic Event Logging) records the perception -> cognition -> action -> physics loop in a tamper-evident hash chain. When a workflow is actually backed by SimulationContract, it can bind geometry hashes, unit validation, deterministic stepping, and replay evidence. Do not treat visual previews as solver evidence unless the contract path emitted the receipt.
Scientific foundation under active verification. The paper program frames trust as composable provenance rather than marketing copy. Current public claims should point to rerunnable evidence, not cached numbers or old pitch decks.
- Trust by Construction: SimulationContract replay, unit checks, and structural convergence evidence; rerun the solver benchmarks before citing a convergence number.
- CAEL Agent Contracts: Hash-chained, replayable, and forkable agent thoughts.
- Trustworthy Tool Use: independent trace replay for MCP tool verification.
- Browser-Native SNN: neuromorphic computing via WebGPU compute shaders (verify benchmarks live).
- Conflict-Free Spatial State: spatial CRDTs resolved via physics-native replay.
- Sandboxed Embodied Simulation: AI-generated physics verified in V8 isolates.
- GraphRAG Self-Understanding: provenance-backed codebase intelligence queries.
Semantic data platform. Any data source → .holo → transform → vector DB + search → compile to dashboard, API, or spatial interface.
Autonomous fleet operations. Swarm intelligence + agent lifecycle + negotiation + economy + board + skills. A fleet that discovers work, negotiates who does it, executes, bills, and reports.
Quick Start (30 seconds)
Try the API — no install needed:
curl -s -X POST https://mcp.holoscript.net/api/compile \
-H "Content-Type: application/json" \
-d '{"code": "composition \"Hello\" { object \"Cube\" { @physics geometry: \"box\" position: [0,1,0] } }", "target": "unity"}' \
| python -m json.tool
PowerShell:
curl -s -X POST https://mcp.holoscript.net/api/compile `
-H "Content-Type: application/json" `
-d '{"code": "composition ""Hello"" { object ""Cube"" { @physics geometry: ""box"" position: [0,1,0] } }", "target": "unity"}' |
ConvertFrom-Json |
ConvertTo-Json -Depth 100
Change "target" to get different platforms from the same input:
| Target | Output |
|---|---|
unity |
C# MonoBehaviour |
r3f |
React Three Fiber JSX |
urdf |
ROS 2 / Gazebo robot XML |
godot |
GDScript scene |
visionos |
RealityKit Swift |
native-2d |
Standalone HTML page |
node-service |
Express.js skeleton |
a2a-agent-card |
A2A Protocol agent manifest |
nir |
Neuromorphic IR (Loihi 2, SpiNNaker) |
All registered targets work the same way — same .holo input, different output. Full list: ExportTarget in packages/core/src/compiler/CircuitBreaker.ts.
See it in a browser:
curl -s -X POST https://mcp.holoscript.net/api/compile \
-H "Content-Type: application/json" \
-d '{"code": "composition \"Store\" { object \"Product\" { @label(text: \"Demo\") @gauge(value: 99, unit: \"%\") geometry: \"box\" } }", "target": "native-2d"}' \
-o demo.html && open demo.html
Scaffold a project:
npx create-holoscript my-world
cd my-world && npm install && npm run dev
Develop on the repo:
git clone https://github.com/brianonbased-dev/HoloScript.git
cd HoloScript && pnpm install && pnpm build && pnpm test
Agent validation (TypeScript): After touching packages/*, run pnpm preflight so only changed packages are typechecked (fast, Windows-safe spawns). Narrow to TS only with pnpm preflight --check=typescript,ts. For dependency posture, use pnpm run health:deps or pnpm preflight --check=dependency_audit; both use the bounded pnpm audit wrapper and emit structured JSON/SKIP instead of hanging. Before a large merge, use pnpm preflight --full (all packages; slower). Flags and checks live in scripts/preflight.mjs.
HoloScript MCP 질문
HoloScript는 스크립트 언어인가요?
아니요, HoloScript는 자체 렉서(lexer), 파서(parser), AST, 타입 및 트레이트(trait) 시스템, VM, 컴파일러 백엔드를 갖춘 범용 시스템 언어입니다.
AI agent가 HoloScript를 사용할 수 있나요?
네, agent는 MCP 도구를 통해 HoloScript를 작성할 수 있으며, 컴파일러는 생성된 결과물이 결정론적(deterministic)임을 보장합니다.
HoloScript는 어떤 타겟으로 컴파일할 수 있나요?
Unity, Unreal, Godot, WebGPU, WASM, iOS 및 URDF/SDF와 같은 로봇 형식을 포함하여 문서화된 25개 이상의 타겟을 지원합니다.
HoloScript를 사용하려면 API key가 필요한가요?
아니요, CLI, 라이브 플레이그라운드를 사용하거나 API key 없이 읽기 전용 공개 도구를 호출할 수 있습니다.