product-cli
MCP server for the Product Framework that enables agents to author and verify a What/How graph, with a live web view for visualizing the domain model, event flows, and system maps.
README
Product
A CLI and MCP server for the Product Framework — specify software as a verifiable What/How graph.
The Product Framework is an open standard for describing a software product as one connected, machine-readable graph: the What (domain model + event model — entities, commands, events, read models, UI steps typed against Abstract Interaction Objects, systems, triggers, Deciders, Projectors), the How (contracts, the screen-composition / reification model, delivery features), and the typed links between them. The graph can drive generation, gate verification, and explain itself — so "describe this system" is a query, not a stale document.
This repo is the reference tooling: a single Rust binary (product) plus an MCP
server that lets an agent author and verify the graph directly. No database, no
service — the graph lives as YAML/Turtle under .product/.
$ product init --demo # scaffold + seed the bookstore What model
$ product domain new system sys-shop --system-kind application \
--purpose "consumer e-commerce" --target-classes gui
$ product domain validate --strict # per-node shapes + graph-level completeness
$ product decider derive Order # derive an aggregate's executable signature
$ product decider validate Order-decider
$ product mcp --http # MCP server + a live Event-Modeling web view at /
Install
# from source
cargo install --path product-cli
The binary ships with the What→How→Build Claude Code skills baked in.
product init writes them into .claude/skills/ of the new repo (pass
--no-skills to opt out); product skills install (re)installs them, and
product skills install --global puts them in ~/.claude/skills/ for every
project. Start a fresh Claude Code session to pick them up, then /product-session.
Choosing the agent CLI
product session start (and product author domain) host the What→How→Build
session in an agent CLI — Claude Code or GitHub Copilot CLI. The CLI is
resolved in this order:
- the
--cli claude|copilotflag, else - the repo's
[author].cliin.product/config.toml, else - the global user default in
$XDG_CONFIG_HOME/product/config.toml(or~/.config/product/config.toml), else claude.
# .product/config.toml — make this repo default to Copilot CLI
[author]
cli = "copilot"
Scaffold it on a new repo with product init --cli copilot, or set a personal
default for every repo by putting the same [author] block in
~/.config/product/config.toml. With a default configured, product session start needs no --cli flag.
60-second tour
product init --demo # a worked What model to explore
product domain list # the captured nodes, by kind
product domain show Order # one node and its links
product domain export # the graph as RDF/Turtle
product domain validate # §3.1/§3.2 per-node conformance shapes
product domain validate --strict # + §3.2.0/§3.2.5/§3.4/§4.5 completeness checks
product decider derive Order # §3.3 — derive decide/evolve signature
product decider simulate Order-decider # run its flow-derived scenarios
product guide # where you are + the next step
The model
- What —
product domain …captures the domain + event model;product decider …(§3.3) andproduct projector …(§3.4) make behaviour and read models executable;product primitive …(§3.5) names irreducible algorithms. - How —
product how,product feature,product build,product seam,product previewcover the How contract, delivery features, the screen seam, and the §11/§12 design-system / content-store preview profiles. - Everything is validated against the framework's SHACL shapes + SPARQL rules;
the captured What serializes to Turtle (
product domain export).
MCP + the web view
product mcp --http starts the MCP server (framework tools: product_domain_*,
product_decider_*, product_projector_*, …) and serves a live web view at /
that renders the active What graph across three connected views — Systems
(the product → systems & journeys map, §3.0), Domain (one bounded context as
an ER graph, §3.1), and Flows (a system's event-model as Event-Modeling
swimlanes — triggers / commands / views over per-aggregate event streams, §3.2).
A node detail panel, the What→How→Build phase stepper, dark/light theme and live
SSE refresh round it out.
Build & test
cargo build
cargo t # full suite (alias: test --no-fail-fast)
cargo clippy -- -D warnings -D clippy::unwrap_used
See CLAUDE.md for the architecture and contributor workflow, and docs/product-framework-open.md for the spec.
License
See LICENSE.
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。
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。
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。
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。