wardrowbe-mcp
Exposes the Wardrowbe wardrobe API as MCP tools, enabling outfit suggestions, item management, wear/wash logging, and analytics via natural language.
README
wardrowbe-mcp
MCP (Model Context Protocol) server exposing the Wardrowbe
wardrobe API as tools an LLM can call. Works against any Wardrowbe instance —
self-hosted (e.g., behind a reverse proxy or the ha-wardrowbe Home Assistant
add-on) or a hosted/cloud deployment.
22 tools covering outfit suggestions, item browsing, wear/wash logging,
acceptance flow, analytics, and notifications. Tool surface mirrors
hacs-wardrowbe's LLM API one-for-one, plus three read-only helpers
(list_items, get_item, get_outfit).
Install
pip install wardrowbe-mcp
Or from source:
pip install git+https://github.com/saya6k/mcp-wardrowbe
Run
wardrowbe-mcp \
--wardrowbe-url https://wardrowbe.example.com \
--api-key "$(openssl rand -hex 32)" \
--auth oidc \
--oidc-issuer-url https://id.example.com \
--oidc-client-id wardrowbe \
--oidc-refresh-token "<refresh-token>"
The process listens on 0.0.0.0:8080 by default and serves both transports
at the root:
http://<host>:8080/mcp— Streamable HTTP (recommended)http://<host>:8080/sse— Server-Sent Events (legacy MCP clients)http://<host>:8080/— anonymous health/info probe
All non-probe routes require Authorization: Bearer <api-key>.
Stdio transport (for proxies)
Bridges like sparfenyuk/mcp-proxy
and the HASS-MCPProxy add-on
spawn MCP servers as child processes and read JSON-RPC frames from their
stdout. Pass --transport stdio (or set MCP_TRANSPORT=stdio) to run in
that mode:
wardrowbe-mcp --transport stdio \
--wardrowbe-url https://wardrowbe.example.com \
--auth oidc --oidc-issuer-url https://id.example.com \
--oidc-client-id wardrowbe --oidc-refresh-token "<refresh-token>"
stdio mode skips Starlette/uvicorn and the Bearer middleware (--host,
--port, --api-key are ignored) — the parent proxy owns the trust
boundary. Backend auth (--auth) still applies because it controls how
the server reaches the wardrowbe backend.
HASS-MCPProxy servers.yaml example
- name: wardrowbe
enabled: true
type: github-python
repo: "https://github.com/saya6k/mcp-wardrowbe"
branch: "main"
install: "uv pip install -e ."
run: "python -m wardrowbe_mcp --transport stdio"
args: []
env:
WARDROWBE_URL: "https://wardrowbe.example.com"
MCP_AUTH_MODE: "oidc"
MCP_OIDC_ISSUER_URL: "https://id.example.com"
MCP_OIDC_CLIENT_ID: "wardrowbe"
MCP_OIDC_REFRESH_TOKEN: "${WARDROWBE_REFRESH_TOKEN}"
After Apply & Restart, the proxy exposes the server at
http://homeassistant.local:8080/servers/wardrowbe/sse — add that URL to
Home Assistant via Settings → Devices & Services → MCP Server.
Configuration reference
Every flag has a matching environment variable so you can drive it from a
shell environment, a .env, or a container orchestrator without rewriting
the command line:
| Flag | Env var | Notes |
|---|---|---|
--transport |
MCP_TRANSPORT |
http (default) or stdio. |
--host |
MCP_BIND_HOST |
Bind address. Default 0.0.0.0. http only. |
--port |
MCP_BIND_PORT |
Bind port. Default 8080. http only. |
--wardrowbe-url |
WARDROWBE_URL |
Base URL of the Wardrowbe backend. |
--api-key |
MCP_API_KEY |
Required Bearer token for incoming MCP calls. http only. |
--auth |
MCP_AUTH_MODE |
dev (default) or oidc. |
--external-id |
MCP_EXTERNAL_ID |
Dev-mode identity sent to /auth/sync. |
--oidc-issuer-url |
MCP_OIDC_ISSUER_URL |
OIDC discovery base. |
--oidc-client-id |
MCP_OIDC_CLIENT_ID |
|
--oidc-client-secret |
MCP_OIDC_CLIENT_SECRET |
Optional for public clients. |
--oidc-refresh-token |
MCP_OIDC_REFRESH_TOKEN |
Obtained from a one-time external PKCE flow. |
--log-level |
MCP_LOG_LEVEL |
DEBUG/INFO/WARNING/ERROR. |
Auth to the Wardrowbe backend
Two modes, selected by --auth:
dev— sends{external_id, email, display_name}toPOST /auth/sync. Works only when the Wardrowbe backend itself is in dev mode (itsDEBUG=trueandSECRET_KEYleft at the upstream default).oidc— refreshes a storedrefresh_tokenagainst the configured issuer to mint a freshid_token, then/auth/sync. The refresh token must come from a one-time interactive PKCE flow you run yourself — there is no in-process browser callback because the MCP server is headless. Once obtained, the server rotates the token automatically if the IDP returns a new one.
get_refresh_token.py — a minimal one-shot PKCE
helper — is the easiest way to bootstrap OIDC. Edit the four constants at
the top (ISSUER, CLIENT_ID, CLIENT_SECRET, REDIRECT_URI), register
that redirect URI on your IDP client, then run:
python3 get_refresh_token.py
A browser opens against the IDP; after you authenticate the script prints
the refresh_token to stdout. Paste it into --oidc-refresh-token (or
the MCP_OIDC_REFRESH_TOKEN env var) when starting wardrowbe-mcp. The
script needs the offline_access scope to be allowed on your IDP client;
otherwise the token endpoint omits refresh_token from the response.
Common IDPs verified: Pocket ID. Should work with any OIDC-conformant
provider (Keycloak, Authentik, Auth0, etc.).
Client config
Claude Code / Claude Desktop
{
"mcpServers": {
"wardrowbe": {
"type": "http",
"url": "http://<host>:8080/mcp",
"headers": { "Authorization": "Bearer <api-key>" }
}
}
}
For SSE-only clients, swap "type": "http" → "type": "sse" and /mcp →
/sse.
Other MCP-aware clients
Any client that supports HTTP transport with a Bearer header works. The
server advertises tools, instructions, and a bundled
agentskills.io-compatible skill at
skill://wardrowbe-skill/SKILL.md (plus sibling resources).
Skill bundle
wardrowbe_mcp/skill/ ships inside the package as the agent-facing
documentation: a SKILL.md manifest with usage guidance and three worked
examples under examples/. The MCP server registers each file as an MCP
resource at startup, so MCP clients that auto-install skills pick it up
automatically. Manual install is also possible by copying the directory
into the client's skill discovery path.
The bundle's prose currently mentions the ha-wardrowbe add-on as the
canonical deployment. If you're running this against a different Wardrowbe
deployment, edit SKILL.md to match.
Development
git clone https://github.com/saya6k/mcp-wardrowbe
cd mcp-wardrowbe
pip install -e .
wardrowbe-mcp --help
Smoke test against a running instance:
curl -fsS http://127.0.0.1:8080/ | jq # anonymous health
curl -fsS http://127.0.0.1:8080/mcp \
-H "Authorization: Bearer $MCP_API_KEY" \
-H "Content-Type: application/json" \
-d '{"jsonrpc":"2.0","id":0,"method":"initialize","params":{"protocolVersion":"2025-11-25","capabilities":{},"clientInfo":{"name":"smoke","version":"0"}}}'
License
MIT. 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。