Skip to main content
Glama
Wardrowbe MCP Server

Wardrowbe MCP Server

by mjrealm

wardrowbe-mcp

An MCP server for Wardrowbe, the self-hosted wardrobe app. It lets any MCP client list and inspect items, log wear and washes, suggest outfits, and check analytics.

Requires Node.js 20 or newer.

Quick start

You need your Wardrowbe URL and an OIDC client for Wardrowbe. Pocket ID works.

  1. In your IdP, create a public client with PKCE. Allow the scopes openid profile email offline_access, and add the redirect URI http://localhost:8765/callback.

  2. Add the server to your MCP client. For Claude Desktop or Claude Code:

    {
      "mcpServers": {
        "wardrowbe": {
          "command": "npx",
          "args": ["-y", "@mdeleon23/wardrowbe-mcp"],
          "env": {
            "WARDROWBE_URL": "https://wardrowbe.example.com",
            "OIDC_ISSUER_URL": "https://pocket.example.com",
            "OIDC_CLIENT_ID": "wardrowbe"
          }
        }
      }
    }
  3. Call list_items from your client. The first time, your browser opens for sign-in. After that, the refresh token is saved and you won't be asked again.

To use a plain access token instead, set WARDROWBE_TOKEN in place of the OIDC_* variables. It expires, so OIDC is better for regular use.

Related MCP server: Pantry Buddy MCP

Configuration

Variable

Required

Default

Purpose

WARDROWBE_URL

yes

Base URL of your Wardrowbe instance.

WARDROWBE_TOKEN

or OIDC

A Wardrowbe access token (JWT).

OIDC_ISSUER_URL

or token

OIDC issuer, e.g. https://pocket.example.com.

OIDC_CLIENT_ID

with OIDC

OIDC client ID.

OIDC_CLIENT_SECRET

no

Only for confidential clients.

OIDC_REFRESH_TOKEN

no

Use this refresh token instead of the saved one. For headless hosts (see below).

WARDROWBE_TOKEN_FILE

no

~/.config/wardrowbe-mcp/refresh-token.json

Where the refresh token is saved.

MCP_TRANSPORT

no

stdio

stdio or http.

MCP_BIND_HOST

no

127.0.0.1

Address for HTTP. Use 0.0.0.0 in containers.

MCP_PORT

no

8080

Port for HTTP.

MCP_API_KEY

http only

Bearer key that clients send to /mcp.

How sign-in works

With OIDC set, the server signs in through your browser the first time a tool needs a token. It catches the redirect on localhost:8765, then saves the refresh token to WARDROWBE_TOKEN_FILE with owner-only permissions. Later runs reuse it. If the token is revoked, the server signs in once more.

The browser sign-in needs a browser on the same machine as the server. On a headless host or in a container, sign in elsewhere and pass the token in:

npx -y @mdeleon23/wardrowbe-mcp login --issuer https://pocket.example.com --client-id wardrowbe > refresh-token.txt

Then set OIDC_REFRESH_TOKEN to the contents of refresh-token.txt, and delete the file.

HTTP transport

For MCPHub or any client that connects to a URL:

MCP_TRANSPORT=http \
MCP_BIND_HOST=0.0.0.0 \
MCP_API_KEY="$(openssl rand -hex 32)" \
WARDROWBE_URL=https://wardrowbe.example.com \
OIDC_ISSUER_URL=https://pocket.example.com \
OIDC_CLIENT_ID=wardrowbe \
OIDC_REFRESH_TOKEN=<refresh-token> \
npx -y @mdeleon23/wardrowbe-mcp

Register the server at http://<host>:8080/mcp with the header Authorization: Bearer <MCP_API_KEY>. GET / returns a small info response without auth.

Adding items from photos

Photos go to the server directly, not through the AI, so the model never has to handle image bytes:

  1. The app uploads the photo to POST /uploads with the same Bearer key. Send the raw image as the body, with Content-Type set to image/jpeg, image/png, image/webp, or image/heic. Optionally set X-Filename (URL-encoded).

  2. The response is {"upload_id": "...", "expires_in_seconds": 600}.

  3. The agent calls add_item with that upload_id. The upload is used once and then deleted. Uploads over 15 MB are refused, and uploads expire after 10 minutes.

curl -s -X POST https://<host>:8080/uploads \
  -H "Authorization: Bearer <MCP_API_KEY>" \
  -H "Content-Type: image/jpeg" \
  -H "X-Filename: blue-shirt.jpg" \
  --data-binary @blue-shirt.jpg

An agent can also do the whole thing through MCP with the upload_photo tool, which takes the photo as base64 (up to 300 KB) and returns an upload_id for add_item. The model has to write the image out in full, so keep photos small for this route. Full-size phone photos should go through /uploads.

add_item works over stdio too, as long as the upload_id came from upload_photo in the same server process. /uploads is HTTP only.

Tools

Tool

Wardrowbe endpoint

Notes

list_items

GET /items

Filter by type, status, favorites, needs wash, search. Paginated.

get_item

GET /items/{id}

log_wear

POST /items/{id}/wear

Defaults to today.

log_wash

POST /items/{id}/wash

Defaults to today.

suggest_outfit

POST /outfits/suggest

Rate limited to about 10 per minute per user.

accept_outfit

POST /outfits/{id}/accept

reject_outfit

POST /outfits/{id}/reject

upload_photo

POST /uploads (in-process)

Takes a small photo as base64 (up to 300 KB) and returns an upload_id.

add_item

POST /items

Takes an upload_id from upload_photo or /uploads. Wardrowbe detects the type unless you pass one.

get_analytics

GET /analytics

days from 7 to 365.

get_wardrobe_summary

GET /analytics

Counts, insights, most and never worn items. Good first call.

get_most_worn_items

GET /analytics

Up to 5 items, from the same trend window.

get_items_to_wash

GET /items?needs_wash=true

Items worn since their last wash.

archive_item

POST /items/{id}/archive

Hides an item from suggestions. Reversible. Takes an optional reason.

restore_item

POST /items/{id}/restore

Brings an archived item back.

Skill

The server publishes skill://wardrowbe-skill/SKILL.md as an MCP resource. It explains which tool to use for each task, how to add an item from a photo (including re-encoding large photos), and how archiving works. Clients that install skills pick it up automatically. The source is in skill/SKILL.md.

Development

npm install
npm run build
npm test

Written against Wardrowbe 1.13.x. If Wardrowbe changes its endpoints, update src/tools.ts.

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections.

Loading tool schema changelog…

    Related MCP Connectors

    Related MCP Servers

    • F
      license
      Not graded
      quality
      B
      maintenance
      Enables MCP-compatible clients like ChatGPT and Claude to manage a household's shared food inventory, meal plans, recipes, and preferences through natural language, with atomic updates and audit logging.
      -
    • F
      license
      Not graded
      quality
      C
      maintenance
      Enables managing a personal pantry, shopping lists, cookbook recipes, and saved receipts remotely through MCP tools using natural language or API calls.
      -
    • A
      license
      Not graded
      quality
      B
      maintenance
      Enables conversational logging and retrieval of workouts, sets, weigh-ins, meals, goals, plans, and progress photos through MCP-capable AI clients, with authenticated per-user storage and export/query capabilities.
      ISC
    • A
      license
      B
      quality
      B
      maintenance
      Enables MCP clients to read account status, plans, and recent workouts, search official exercise names, and, when writes are enabled, apply confirmed preview-first plan changes and historical workout logs.
      22
      MIT