text-to-reality

Describe something you want to exist. Your agent turns it into a build package anyone can put together by following pictures:

• a parts list you can order, using parts that plug together
• 3D-printable parts for the case and mounts
• the wiring, every wire and which connector it goes in
• the code, ready to load
• a step-by-step picture guide written for someone who has never built electronics
• tests, so you know it works

No soldering, no breadboard, no electronics experience required. It runs on your machine: no account, no API key, and your projects are plain folders you own.

Example: Agent Meter

"Build me a little desk character that shows what my coding agent is doing and how much of my Claude allowance is left."

[Agent Meter, a small printed robot character with a screen in its head (concept render)]

Agent Meter illustrates this workflow: a Seeed Studio XIAO ESP32-S3 and a small screen inside a printed character, wired with plug-on connectors, with firmware and a picture guide. Its face shows when your agent is working, finished or waiting on you, plus your remaining allowance.

Get the Agent Meter kit →

Install

Ask your agent

Install text-to-reality from https://github.com/ai-wes/text-to-reality

Or install it yourself. text-to-reality runs through uv: check it's installed with uv --version, or use uv's installer. Then run the commands for your agent and restart it.

Claude Code

claude plugin marketplace add ai-wes/text-to-realityclaude plugin install text-to-reality@ai-wes

To update: claude plugin marketplace update ai-wes, then claude plugin update text-to-reality@ai-wes.

Codex

codex plugin marketplace add ai-wes/text-to-realitycodex plugin add text-to-reality@ai-wes

Cursor

Cursor also loads the Claude Code plugin; if that's installed, skip this.

git clone --depth 1 https://github.com/ai-wes/text-to-reality ~/.cursor/plugins/local/text-to-reality

Claude Desktop

Add the server to Claude Desktop's config (Settings > Developer > Edit Config), then restart. If it can't find uvx, use its full path (which uvx).

{  "mcpServers": {    "text-to-reality": {      "command": "uvx",      "args": ["--no-config", "--managed-python", "--python", "3.13", "--from", "text-to-reality==0.1.1", "text-to-reality", "mcp"]    }  }}

Projects are saved in text-to-reality/ in the folder the app starts in; set TEXT_TO_REALITY_DIR in env to choose another folder.

Other agents

npx skills add ai-wes/text-to-reality

Then add the MCP server above to your agent's MCP config.

Works with text-to-cad

text-to-reality plans the whole device; text-to-cad is great at the printed parts. Install both, and the agent uses text-to-cad for the printed parts when it's available.

How it works

Each build is a folder. The agent fills in seven stages and checks them:

The local MCP server gives the agent tools to create projects, record files, catch files edited after they were checked, validate the package, keep a build history, and look up board header layouts and plug-on parts.

You can also check a folder yourself:

uvx text-to-reality check ./text-to-reality/my-build

Parts that plug together

Loose jumper wires fall off and get swapped. JIG_ makes connectors that turn the wires going to one row of pins into a single plug, so wiring becomes plugging things in.

The agent uses JIG_ connectors by default. It looks up each board's pin rows, picks one connector per row, lists what to buy, and notes any row a connector can't fit. Get them at jig-robotics.com →

Each connector is a base that holds the wires in order and a cover that snaps on. The picture guide shows how to load it, close it and plug it on.

License

MIT