glasz-mcp
Fetches glasz-style spec files (product.md, tech.md, test-plan.md, and INDEX.md) from GitHub repositories, enabling LLMs to answer questions about features, stories, and acceptance criteria without cloning or pasting files.
README
glasz-mcp
A read-only Model Context Protocol server that
fetches glasz-style spec files (the product.md / tech.md / test-plan.md
trio + the derived specs/INDEX.md) from GitHub repositories and hands them to
an LLM client, so the model can answer questions about features, stories, and AC
status without the user cloning the repo or pasting files.
It is intentionally a thin window onto the same source data that glasz the app
reads — no writes, no parsing on the server (the LLM interprets the trio,
guided by the server instructions block).
Status — v1-core
This release implements the v1-core scope:
| Story | What |
|---|---|
| S1 | Install via npx -y github:quintegro/glasz-mcp (compiled dist/ is committed) |
| S4 | get_index — fetch specs/INDEX.md |
| S5 | get_feature — fetch a feature's trio |
| S6 | instructions schema legend taught at initialize |
Deferred to v1.x: S2 (OAuth device-flow login — currently a stub) and
S3 (list_repos discovery — not registered).
Install
Add this to your MCP client config (Claude Desktop / Claude Code / Cursor):
{
"command": "npx",
"args": ["-y", "github:quintegro/glasz-mcp"],
"env": { "GITHUB_REPO": "owner/name" }
}
GITHUB_REPO is optional — when set, tools default to it if repo is omitted.
Authentication
Public repositories are readable without a token. For private repos, set
GITHUB_TOKEN in the client config env, or place a token at
~/.glasz-mcp/auth.json (mode 0600):
{ "access_token": "ghp_…", "token_type": "bearer", "scope": "repo", "login": "you", "created_at": "2026-06-19T00:00:00Z" }
The OAuth device-flow login command is deferred to v1.x.
Tools
get_index({ repo? })— returns the raw markdown ofspecs/INDEX.md.get_feature({ repo?, slug })— returns the trio forspecs/features/<slug>/.
Development
pnpm install # also runs `prepare` → build
pnpm run build # tsc → dist/
pnpm test # unit + integration (vitest)
node bin/glasz-mcp.js # start the server on stdio
dist/ is committed to git (not .gitignored) because npx git-installs do
not run build scripts — the compiled output must exist on the default branch.
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。