# Metatable Features | From Plain-English Plan to Running App

> One plan. A complete, running application built from it.

You describe how your business works. Metatable writes it down as a plan you can read, then builds everything from it — the database, the rules, the screens, the hosting. Change the plan and the software follows. Export the code anytime.

## 1. The plan

It starts with a plan you can read — not a pile of prompts. Before anything is built, Metatable writes down your roles, records, rules, and edge cases in plain language. You read it, correct it, approve it. The app is built from that document — and when your business changes, you change the document, not the code.

For the technically minded: it is a real spec, and it stays the source of truth.

A typical approved plan covers:

- **Product overview** — replace spreadsheet-based vendor tracking with a running application.
- **Primary workflow** — capture request, assign owner, validate data, approve, report status.
- **Access rules** — editors update operational records; owners manage users and approvals.
- **Edge cases** — duplicate supplier, expired contract, missing approver, delayed renewal.

## 2. What you get

A complete, running application — typically live in about 30 minutes. Not screens over a spreadsheet: every layer of real software, plus the artifacts that make it easy to review, change, and take with you.

1. **A real database** — every record your business tracks, stored safely with proper relationships. (PostgreSQL — the same class of database banks run on.)
2. **The engine that runs your rules** — approvals, permissions, thresholds, calculations — enforced automatically, exactly as the plan says. (Rust backend with typed APIs.)
3. **Screens for your team** — pages to enter, find, and approve things, built from the workflows and roles in your plan. (Flutter web app.)
4. **Hosting, handled** — live on managed infrastructure from day one. Export the full source code whenever you want. (Azure. No lock-in.)
5. **Your plan, kept current** — readable documentation that always matches what is actually running.
6. **AI assistant access** — use the whole thing from ChatGPT or Claude — no screen required. (MCP server — see below.)

## 3. Use it anywhere

Use your app from the AI assistant you already have. Ask questions, pull reports, even trigger workflows from ChatGPT or Claude — no dashboard, no new login. Under the hood, your app exposes an MCP server: the standard that lets AI assistants use software as tools.

Private by default: it is your data in your account, and you choose which services are exposed and what they are allowed to do.

Example:

- **You:** What is below reorder threshold right now?
- **Tool call:** called inventory backend
- **Assistant:** 3 SKUs are below threshold: Blue Label 750ml has 18 left, reorder at 40; Citrus Mixer 12-pack has 22 left, reorder at 50; Gift Box Medium has 9 left, reorder at 25.
- **Follow-up:** Want me to create the purchase orders? Owner approval is required.

Read more: [Software You Talk To](/blog/software-you-talk-to-headless-mcp)

## 4. Why it holds up

Why it doesn't fall apart at 80%: turning a plan into working software is not one prompt. It is a pipeline that builds in the right order, tests against the real thing, and fixes its own errors before you ever see the build.

**87% of backend builds succeeded over the last 90 days — 100% in the last 30.**

1. **It builds in the right order** — tables before APIs before screens, following the actual dependencies — not whatever the model happens to emit first. (DAG-ordered generation.)
2. **It tests against the real thing** — every query runs on a real database engine and every line of backend code is actually compiled. Errors are found by running the software, not guessed from static analysis. (Execution-based validation.)
3. **It fixes its own mistakes** — the system reads the real compiler and database errors, corrects its output, and runs it again until the build holds. (Self-fixing loops.)
4. **It doesn't lose the plot as your app grows** — each step sees only the relevant part of your plan, so the tenth change is as reliable as the first. (Context engineering.)

## 5. Honest fit

Built for business software, not every possible app.

- **Best fit** — CRMs, internal tools, inventory, portals, trackers, approval flows, and operational dashboards.
- **Use with review** — customer-facing workflows where the data model, permissions, and handoff matter more than visual novelty.
- **Not best yet** — highly custom consumer interfaces, games, or products where frontend polish is the main product.
- **Honest limit** — frontend generation is still maturing. The strongest fit today is structured business software.

## Start here

Have a workflow that should become software? Describe it, read the plan, and see whether Metatable can turn it into a running application.

- Build my app free: https://app.metatable.ai/signup
- Book a 15-min demo: https://cal.com/metatable/ai-dev-tool-demo