Bundlekit
A standalone local MCP server for managing docs-hub bundle IRs, including splitting, merging, checking, normalizing bundles, rendering design Markdown, validating legacy dynamics YAML, and syncing owned docs skills.
README
Bundlekit
Standalone local MCP for docs-hub bundle IR:
- split / merge / check / normalize bundles
- render design Markdown
- validate legacy-dynamics YAML
- sync owned docs skills on
init
Package is independent of Hubdocs, ArtifactGraph, and CodeGraph. They remain
optional accelerators; missing optionals use the documented fallback and
measurable bundlekit.missing-optional event contract.
Install
# from a local checkout (dev)
cd /path/to/bundlekit
pnpm install
pnpm build
pnpm link --global # optional
cd /path/to/docs-hub
bundlekit init
Linux installer (after the repo is published):
curl -fsSL https://raw.githubusercontent.com/raintr91/bundlekit/main/install.sh | bash
Init
cd <docs-hub>
bundlekit init
The TTY wizard runs in three steps:
- choose agents (detected agents are pre-selected; choosing none is valid)
- optionally configure installed Hubdocs and/or ArtifactGraph toolkits (none means skip and add them later)
- when Cursor and machine-local repo maps are present, wire cross-repo
CodeGraph servers now or defer to
platform-dna codegraph:wire
Bundlekit never downloads another toolkit implicitly. A selected optional toolkit that is not installed is reported as deferred without failing init. CodeGraph wiring is owned and performed by Platform DNA, not reimplemented inside Bundlekit.
Writes/updates:
- project-local MCP configs for the agents selected by the wizard
- owned skills/extracts/rules under
.cursor/(commit-worthy; not ignored wholesale) - exact machine-local ignore entries (
.bundlekit/,legacy-repos.local.json,platform-repos.local.json, written agent MCP paths such as.cursor/mcp.json) - both machine-local map skeletons when missing (never overwrite member content)
- repo-only
legacy-repos.json+ example for optional brownfield evidence - Bundlekit CLI aliases in
package.jsonwithout replacing existing scripts - no portable
platform-repos*.json: that config is Platform DNA-owned
Configure repo maps (/configure-repo-maps)
After init, fill machine-local checkouts with natural language — do not hand-edit
JSON. Prefer the Platform DNA skill when that toolkit is installed; Bundlekit
ships a thin copy for independence. /legacy-spec Gaps when a legacy root is
missing hand off here.
Platform example (writes platform-repos.local.json):
docs = base-docs ở ~/ws/base-docs; portal admin ở ~/ws/portal; api core ở ~/ws/api
Legacy example (writes legacy-repos.local.json, key legacy-*):
legacy ERP cũ ở D:\legacy\erp, key legacy-erp
For CI, use bundlekit init --yes; detected agents and the only supported lane
(docs) are selected automatically, optional toolkits and CodeGraph are skipped
by default. Use --with=hubdocs,artifactgraph (or --with=none) and
--codegraph|--no-codegraph for explicit automation. --target, --type,
--location, and --project-root remain available.
Tools
| Tool | Purpose |
|---|---|
bundle_split |
*.bundle.yaml → ir/* |
bundle_merge |
ir/* → bundle |
bundle_check |
split drift check |
bundle_split_all |
batch split |
bundle_normalize |
normalize gen |
docs_render / docs_render_common |
design MD only |
legacy_dynamics_validate |
legacy dynamics schema |
Owned skills
/spec · /update-spec · /update-spec-legacy · /legacy-spec ·
/configure-repo-maps · /bqa-grill-docs · /dev-grill-docs · /grill-with-docs
Lifecycle
The primary commands are:
bundlekit init # install in this docs hub
bundlekit deinit # remove this hub's owned harness + local MCP
bundlekit uninstall # preview global/all removal from anywhere
bundlekit uninstall --yes # all tracked hubs + local/global MCP + CLI
bundlekit prune # dry-run
bundlekit prune --yes # delete only unmodified stale files
Interactive deinit/uninstall preview and confirm. Non-TTY runs are dry-runs
unless --yes is passed. init records each destination in the XDG state
ledger and records managed-file hashes in .bundlekit/install-manifest.json.
Use bundlekit uninstall --discover <dir> to find installs created before the
ledger existed.
Customized files and package scripts are preserved and reported. Bundlekit removes
legacy-repos.json only when it created the file and its recorded hash is
unchanged. Shared MCP/extract registries are unmerged without deleting other
toolkits' entries. prune remains stale-only.
See mcp-package.json for tool/harness API compatibility.
Docs
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。