Bezalel

“I have filled him with the Spirit of God, with skill, ability, and knowledge in all kinds of craftsmanship” (Exodus 31:3). Bezalel was the craftsman who made the tabernacle. This Bezalel is a small, open MCP server and a matching skill: they let the AI assistant your team already uses, such as Claude, Cursor or ChatGPT, draw on this reference while you design. It is built to slow you down at the right moments. It brings the whole spectrum of options, the research with its limits and the questions to ask, and it leaves the deciding to you.

Why Bezalel

When Israel built the tabernacle in the wilderness, God gave Moses a pattern: “You must make the tabernacle and design all its furnishings according to the pattern I show you” (Exodus 25:9). The work went to Bezalel son of Uri, of the tribe of Judah, filled “with skill, ability, and knowledge in all kinds of craftsmanship” (Exodus 31:3). He did not work alone. Oholiab of the tribe of Dan worked beside him, and the two were given “the ability to teach others” (Exodus 35:34).

Three things in that story shape this tool. Making begins with a pattern received, and it still needs skilled hands and careful judgment. Skill is given to be passed on. And when the people kept bringing gifts, the craftsmen told Moses they had “more than enough for doing the work” (Exodus 36:5), and the people were asked to stop. A good tool knows what is enough.

The image follows the same line. Silicon is ordinary stone, and it becomes the chips in every phone. Here a raw piece has one face polished into a wafer, and the light it catches is not its own.

Three ways in

Pick the one that fits the tool your team already uses. All three carry the same content and the same stance.

• Connector (MCP, hosted): the address below, added once to Claude, ChatGPT, Cursor or another assistant. Always current; works in web and desktop apps without anything installed. Start here.
• Desktop bundle (MCP, on your computer): bezalel.mcpb. Open it with Claude Desktop to install. It runs on your own machine with a snapshot of the reference, so it works offline and nothing leaves your computer. Download it again for newer content.
• Skill: bezalel-skill.zip. Instructions your assistant loads only when a faith-product design question comes up: how to run a review, what to show and what to ask first. Unzip it into ~/.claude/skills/ for Claude Code, upload the zip under Skills in the Claude apps, or add it to any tool that reads SKILL.md. It uses the connector when connected, the content API otherwise, and its own index offline.

Connector, skill or command line?

A connector (an MCP server) gives the assistant tools to call: it can look things up in the reference whenever it needs to. It works wherever MCP is supported, including apps with no access to your files. A skill gives the assistant a way of working: a short set of instructions it reads only when the task calls for it, so it costs almost nothing until then. Skills need a tool that supports them, such as Claude Code or the Claude apps. The two complement each other: the skill carries the method, the connector the content. A separate command-line tool would add little: the reference is already plain JSON in the content API, which any script can read with curl.

Add the connector

The server address is https://patterns.missional.design/mcp. No account or key is needed.

• Claude (web and desktop): add a custom connector with this address in your connector settings.
• Claude Code: claude mcp add --transport http bezalel https://patterns.missional.design/mcp
• Cursor and other editors: add an MCP server of type “streamable HTTP” with this address to your MCP configuration.
• ChatGPT: where your plan allows custom connectors, add this address as an MCP connector.

Connector menus change often; if your tool asks for a transport, choose Streamable HTTP.

What to ask

Describe a feature the way your team would, then ask for a review. For example: “Use Bezalel to review a daily reading streak with push reminders,” or start one of its prompts: Design review, Three ways debate or Pre-mortem: who could this hurt?

What it can do

• map_feature: the decisions inside a feature, each with its full spectrum of options (including leaving it out), their provenance, and the questions to ask.
• challenge_default: for a feature copied from mainstream apps, what that pattern is built to maximise, what the research adds, and alternatives that serve the person.
• three_ways: the exploitative, ethical and redemptive versions of a feature, for a debate rather than a verdict.
• premortem: the most vulnerable people likely to use it, and the question to ask for each.
• draft_worksheet: the decision worksheet, with pointers for each step, for your team to fill in.
• search_reference, list_deep_dives, get_deep_dive, find_sources and principles: the reference itself, with every source’s limits.

How it behaves

Bezalel offers options, evidence and questions, not verdicts. It asks your assistant to show every option, to ask your team before recommending, to link every page it draws on, and to leave theological, pastoral and safeguarding judgments to people who can make them. There is no AI on the server: your own assistant does the reasoning with what Bezalel returns.

Privacy

The desktop bundle and the skill run in your own assistant and send nothing to us; the skill fetches public pages only when it has no offline answer. The hosted connector keeps nothing. Each request is answered and forgotten; the feature descriptions you send are not stored or logged, and there are no accounts or cookies. The web server keeps standard access logs (time, address and path, not content) for security.

Licence

The content, including the snapshot in the bundle and the skill’s references, is CC BY 4.0 and the server code MIT. The same content is available as plain data in the content API. Bezalel is new: please tell us how it works in your team’s hands: which assistant you used, what you asked, what helped and what it got wrong.