mcp-features

mcp-features

Provides Dev Container Features that install code intelligence (LSP) and repository knowledge search (Orama) MCP servers into any dev container, enabling coding agents to perform go-to-definition, find references, and hybrid search over project files.

Category
访问服务器

README

mcp-features

Reusable Dev Container Features that drop MCP servers into any dev container, giving coding agents (Claude Code and any other MCP client) two capabilities:

Feature What it gives the agent Transport Package
lsp-mcp Code intelligence — go-to-definition, find-references, hover/types, document & workspace symbols, diagnostics — backed by real language servers. HTTP (warm, long-running service) packages/lsp-mcp
orama-mcp Repo knowledge search — hybrid (BM25 + vector) search over every Markdown and JSONL file in the workspace, kept live as files change. stdio (client-spawned) packages/orama-mcp

Both are MIT licensed, published as public OCI artifacts under ghcr.io/quebi-gmbh/..., and installable with one line in devcontainer.json.


Design philosophy

These features are the "local, fresh, zero-infra" half of a code-and-knowledge retrieval system. The guiding rule: everything is a pure function of the local clone — no external database, no sync service, no quota. Two tiers:

Tier 1 — live tools, no index, no embeddings. The agent already has (or gets, via lsp-mcp):

  • ripgrep — lexical code search
  • ast-grep — structural code search
  • lsp-mcp — semantic code (defs / refs / types / diagnostics)
  • git CLI (incl. git log --grep / -S / -G) — history, blame, diff, commit search
  • gh CLI — specific PR / issue lookup, on demand

Tier 2 — the only thing that embeds, over prose only. That's orama-mcp: Markdown docs and JSONL knowledge records, indexed in-memory and updated live.

Deliberately out of scope: vector-indexing code or git (the CLI/LSP tools above are exact and sufficient), and any always-running GitHub mirror (use the gh CLI on demand, or a periodic batch dump to committed JSONL that orama-mcp then indexes).

See each package README for the detailed tool surface and build plan.


Usage (once published)

Add to a repo's .devcontainer/devcontainer.json:

{
  "features": {
    "ghcr.io/quebi-gmbh/mcp-features/lsp-mcp:0": {
      "languages": "typescript,python"
    },
    "ghcr.io/quebi-gmbh/mcp-features/orama-mcp:0": {
      "globs": "**/*.md,**/*.jsonl"
    }
  }
}

Rebuild the container and the tools are installed, started, and (optionally) registered with any MCP client in the workspace.


MCP registration contract

Each feature can register itself by writing/merging a .mcp.json at the workspace root (the format Claude Code auto-discovers). This is controlled by the autoRegister option (default true).

// .mcp.json (what the features produce)
{
  "mcpServers": {
    "orama": {
      "command": "orama-mcp",
      "args": ["--globs", "**/*.md,**/*.jsonl"]
    },
    "lsp": {
      "type": "http",
      "url": "http://127.0.0.1:7337/mcp"
    }
  }
}

claude-manager integration. claude-manager may prefer to own .mcp.json itself. In that case set autoRegister: false on both features and have the daemon write the two entries above into the project .mcp.json before launching the Claude session. The registration contract (stdio command/args for orama, http url for lsp) is stable — that's the only coupling between this repo and claude-manager.


Repository layout

mcp-features/
├── packages/            # the MCP server implementations (Bun + TypeScript)
│   ├── lsp-mcp/         #   HTTP MCP service wrapping language servers
│   └── orama-mcp/       #   stdio MCP server over Orama
├── src/                 # the Dev Container Features (what ghcr publishes)
│   ├── lsp-mcp/         #   devcontainer-feature.json + install.sh
│   └── orama-mcp/
├── test/                # devcontainers CLI feature tests
├── turbo.json           # Turborepo pipeline
└── package.json         # Bun workspaces root

Why the packages/src/ split? packages/* holds the actual server code (built/tested with Bun + Turbo). src/* holds the Dev Container Features, whose install.sh fetches the corresponding published server (npm or a release binary) and wires it up. The devcontainers publish/test GitHub Actions operate on src/.


Development

Requires Bun ≥ 1.2.

bun install
bun run build       # turbo: build all packages
bun run typecheck
bun run test

Test a feature locally with the devcontainers CLI:

npm i -g @devcontainers/cli
devcontainer features test --features orama-mcp --base-image mcr.microsoft.com/devcontainers/base:ubuntu .

Publishing

Features are published as OCI artifacts to ghcr.io/quebi-gmbh/mcp-features/<feature-id> by .github/workflows/release.yml (the official devcontainers/action). Bump the version in each feature's devcontainer-feature.json and push to main.


Status

🚧 Scaffold only. This repo currently contains the shell, config, and the build plan in each README. Nothing is implemented yet — see the ## Build plan / TODO sections in each package and feature README for what to do next.

推荐服务器

Baidu Map

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。

官方
精选
JavaScript
Playwright MCP Server

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。

官方
精选
TypeScript
Magic Component Platform (MCP)

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。

官方
精选
本地
TypeScript
Audiense Insights MCP Server

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。

官方
精选
本地
TypeScript
VeyraX

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。

官方
精选
本地
graphlit-mcp-server

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。

官方
精选
TypeScript
Kagi MCP Server

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。

官方
精选
Python
e2b-mcp-server

e2b-mcp-server

使用 MCP 通过 e2b 运行代码。

官方
精选
Neon MCP Server

Neon MCP Server

用于与 Neon 管理 API 和数据库交互的 MCP 服务器

官方
精选
Exa MCP Server

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。

官方
精选