The hard part of starting out is not the painting. It is choosing paints and
knowing what you lack. Give any agent your whole shelf and it works from what you actually
own, not a random set that changes every time you ask.
Plan
A scheme from your own shelf
Drop in whatever you have: one photo, two, or a half-painted mini next to the
reference you are chasing. Agents are smart, and now they can see your shelf, so they
work the scheme zone by zone and ground every pick in the pots you already own. Where
there is a real gap they name one or two worth buying, never a fresh set.
Without it: a correct but generic answer that changes every time you ask.
Restock
Log paints without typing
Snap your paint rack or paste a screenshot of your last order. minipainter recognizes
the products and adds them to your inventory. No names to type, no catalog to build by
hand. Your shelf stays current with almost no effort.
One photo in, your inventory back out.
A real sessionParallax (Hal Jordan)
Drop the mini and a reference into the chat.
The agent maps every zone to paints you own, and names the one honest gap worth buying.
Pick a color. Find the paint.
Every paint carries an approximate RGB. Drop a target and minipainter
returns the nearest pot in each brand by straight color distance. Flip to the example
rack below and its paints float to the top. In your install, that rack is whatever you own.
#C9A24C
1,607 pots, one wall
The full catalog, every pot a real swatch. Hover any chip to read it.
The green corner marks a paint in the example rack (a generic sample, not a real inventory).
Hover a chip to inspect a paint.
Hand it to any agent
minipainter speaks the Model Context Protocol, so any MCP-capable agent can
use it: Claude on desktop, web and Code, ChatGPT, Gemini, and whatever comes next. Run it
locally or host your own, there is no shared service. Full setup
guide for each.
Local · CLI + MCP
On your machine
The simplest path, nothing to host. Use minipainter as a CLI in your terminal, or run
the local MCP server and add it to a desktop agent like Claude Desktop. Your rack never
leaves your machine.
For URL-based agents (Claude web, ChatGPT, Gemini), deploy your own remote MCP with
Docker or Fly, then add your URL as a connector. It is your server and your inventory,
there is no public instance to share. ChatGPT uses the /mcp/v3 route.
https://your-app.fly.dev/mcp
paint_searchmatch_colorinventory_listsearchfetch
Agent runtime · skill
Agent Skill
A standards-compliant Agent Skill for runtimes that load skills (Claude Code). It
teaches the agent to drive the CLI with the right guardrails: owned-first, cross-brand
product_format, and ambiguity checks before any write. Fetch it straight
from this site, no clone required.
paint searchpaint showmatch colormatch describeinventory owninventory list
Quickstart
# straight from npm, no clone$ npx minipainter paint search bone
$ npx minipainter paint search black --json
# track what's on your rack$ npx minipainter inventory own "Abaddon Black"
$ npx minipainter inventory list
# match by color or description$ npx minipainter match color "#d2c29b"
$ npx minipainter match describe bone
# open the workshop ledger TUI$ npx minipainter tui