MASA
Provides a local, read-oriented MCP server for the MASA protocol, enabling offline validation, inspection, lineage tracing, public-export audit, and processing-request planning for sound-matter records.
README
MASA
MASA is a project-agnostic Sound Matter Aware protocol. It provides a local-first, language-neutral contract for systems that need to describe, listen to, measure, transform, granulate, dissect, generate, map, relate, govern, and exchange sound as matter without confusing an audio file, a physical event, a perceptual account, or a computational claim. Conformance never depends on any other repository.
Version 0.1.0 is the first public release, MIT-licensed and open to any project. Its normative sources are the specification, JSON Schemas, ontology registries, and conformance fixtures in this repository. The TypeScript packages, CLI, and MCP server are reference implementations of those artifacts.
What is included
- a compact Core plus Audio, Listening, Analysis, Transformation, Generation, Processing, Mapping, Agent, and Publication profiles;
- JSON Schema 2020-12 contracts with stable versioned identifiers;
- an engine-neutral processing-request contract for granular, spectral, and time-pitch operations, with the granular lexicon carried as attributed vocabulary;
- provenance-aware terminology and relation registries;
- a JSON-LD context and documented PROV-O and ODRL mapping boundaries;
.masa.json, directory-bundle, and.masa.zipexchange forms;- offline structural, semantic, lineage, policy, integrity, and public-disclosure validation;
- deterministic local bundle packing and bounded extraction;
- browser-safe TypeScript model and validation packages;
- Node-only bundle tooling and the
masacommand-line interface; - a read-oriented local MCP stdio server;
- conformance fixtures for Reader, Writer, Transformer, Agent Host, and Publisher behavior;
- local skills and non-normative adapter guidance for concrete integrations.
MASA is not an audio engine, model, hosted registry, universal ontology of sound, or claim that machines hear like humans or other species. It can be implemented by conventional software without agents.
Quick start
Requirements: Node.js 22.20 or newer and pnpm 10.32.1.
pnpm install
pnpm check
pnpm masa -- validate examples/0.1.0/valid/minimal-record.masa.json
pnpm masa -- inspect examples/0.1.0/bundles/transformation.masa
pnpm masa -- lineage examples/0.1.0/valid/processing.masa.json urn:uuid:00000000-0000-4000-8000-000000000d02
pnpm masa -- process template granulate
pnpm masa -- conformance reader examples/0.1.0/valid/minimal-record.masa.json
Start the local MCP server over stdio:
pnpm build
pnpm mcp
The MCP server is offline and read-oriented by default. It exposes specification resources and local validation, inspection, lineage tracing, public-export audit, and processing-request planning tools inside configured roots. It does not fetch remote resources, perform signal processing, generate audio, publish, delete, or expose raw assets.
See docs/project-agnostic-integration.md for the portable adoption boundary and minimal implementation sequence, and docs/processing.md for how granular, spectral, and time-pitch work is planned, delegated to external engines, and accounted.
Local conformance evidence
The reference implementation records separate evidence for the reader, writer, transformer, agent-host, and publisher classes in conformance/0.1.0/evidence. The current local run reports all five classes conformant on Node.js 22.22.3 for macOS arm64 against fixture-manifest digest 0240f5403d9d626c3eef75a55eb640cea84cf5b6d17d2824fe4874561c27be95, computed over the examples, capability catalog, schemas, ontology registries, and JSON-LD context.
Run pnpm check to regenerate derived code and verify conformance on the current machine. Run pnpm conformance:evidence only when intentionally recording a new local evidence snapshot. Conformance is always specific to an implementation version, MASA version, profile, class, platform, and exact fixture set; it is not a universal certification.
Repository map
spec/ normative protocol and profile documents
schemas/ canonical JSON Schema 2020-12 contracts
ontology/ canonical terminology and relation registries
contexts/ optional JSON-LD context
examples/ valid, invalid, and bundle examples
conformance/ executable fixtures, runner, and evidence
packages/core/ browser-safe types, IDs, vocabulary, projections
packages/validator/ browser-safe offline validation
packages/bundle/ Node-only directory and ZIP bundle tooling
cli/ local `masa` command
mcp/ local stdio MCP reference server
skills/ bounded operational procedures
adapters/ application adapter contracts
rfcs/ protocol change process
Normative boundaries
The protocol distinguishes:
- physical event, signal, representation, and MatterRecord;
- source origin and causal attribution;
- heard, measured, inferred, interpreted, speculative, and undetermined claims;
- confidence, source health, freshness, actor, and disclosure;
- observation, derived relation, mapping, synthesis, performance, and speculation;
- copy, encoding derivative, continuation, inpainting, variation, mutation, and graft;
- lineage, authorship, ownership, consent, and ethical legitimacy.
Unknown or protected facts are represented explicitly rather than by empty strings. Technical access does not grant permission. A mapping does not become the source's literal voice.
Status and integration
This repository is self-contained and offline-first. No hosted service, public registry, package publication, or concrete application integration is claimed here. Application integrations begin only after this repository's local conformance evidence passes, live in their own repositories, and are versioned independently. The adapter documents under adapters/ are illustrative archetypes, not integration claims.
The reference architecture lives in docs/reference-architecture.md, and the portable adoption boundary lives in docs/project-agnostic-integration.md.
License
MASA code, schemas, specification, and examples are licensed under the MIT License unless a file states otherwise.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。