> For the complete documentation index, see [llms.txt](https://docs.buzzy.buzz/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.buzzy.buzz/the-building-blocks/mcp/buzzy-builder-mcp.md).

# Buzzy Builder MCP

Buzzy Builder MCP is the recommended workflow for using an AI agent to build or change a Buzzy app. The agent connects to Buzzy's MCP server, follows Buzzy's bootstrap instructions, and works with you through staged app-building and review.

Use this section when you want an agent to create, inspect, or edit the app itself. If you already have an app and want that app to expose tools to ChatGPT or Claude, use [Buzzy Custom MCP](/the-building-blocks/mcp/buzzy-custom-mcp.md) instead.

{% hint style="success" %}
**Why this is the recommended AI creation path:** in our experience, Builder MCP has often been the most cost-effective path and produced the best-quality results for serious app creation and editing, while supporting staged review and control. Use it through an MCP-capable agentic AI development tool such as Codex, Claude Code, Cursor, or Antigravity. The connected tool's subscription, usage allowance, billing, and terms apply; Buzzy Deployment subscriptions do not include a recurring monthly AI credit allowance.
{% endhint %}

Results depend on what you are building, the instructions you provide, the selected model, and current model performance. AI can make mistakes. Review and test generated app artifacts, permissions, screens, code, and runtime behavior before publishing. See [AI usage and pricing](/working-with-buzzy/buzzy-ai/ai-tokens-and-pricing.md) for all three AI-assisted creation paths.

## Watch the overview

This two-minute walkthrough shows how to start with a Buzzy template and adapt it with Claude Code, Codex, Cursor or another compatible agent.

{% embed url="<https://www.youtube.com/watch?v=asPvKNbCxv4>" %}
Build from templates with Buzzy Builder MCP
{% endembed %}

### Start with a pattern, then adapt the app definition

In the [ShortStay-to-CareConnect example](https://www.buzzy.buzz/post/short-stay-to-care-connect-buzzy-builder-mcp/), property listings become carer profiles and nightly stays become hourly visits. The booking pattern provides a starting point; the agent adapts the data, roles, screens and workflows for the new application.

The agent edits the structured app definition while Buzzy's maintained engine runs it. This reduces the custom application code your team needs to own. You still maintain your requirements, access rules, integrations and any custom extensions, and test important user journeys before release.

The same approach can support other applications when the starting pattern fits: tutoring bookings, equipment hire or a service marketplace from a booking pattern, or a sales workflow from a CRM template. Check the template's actual capabilities and adapt missing business rules in stages.

### Put your existing agent allowance to work

Direct artifact edits can use an eligible agent subscription's included allowance when you sign in through that subscription. Claude Code and Codex offer subscription sign-in paths; availability, plan limits and terms apply. API-key authentication is a separate billing path. Check your active account and extra-usage settings before starting.

Direct adaptation does not require Buzzy generation credits. Buzzy-managed generation and deployment have separate costs, as explained below.

### Try the workflow

1. Choose a close starting point from [Buzzy templates](https://app.buzzy.buzz/editor/templates).
2. Follow [Getting started](/the-building-blocks/mcp/buzzy-builder-mcp/getting-started.md) to connect your agent and bootstrap its workspace.
3. Give the agent the template's editor URL and describe what must change. Ask it to create your own target app and preserve the source template.
4. Review the proposed data, roles and journeys, then test the result in preview or runtime.

For a complete worked example, follow the [carer app tutorial](/the-building-blocks/mcp/buzzy-builder-mcp/tutorial-carer-app-from-a-template.md).

## Create vs Adapt costs

* **Create from scratch:** Buzzy-managed generation of a new brief, flows, data model, theme, blueprint, sample data, or screens requires separately purchased Buzzy AI credits or a supported bring-your-own OpenAI API key, in addition to the agentic tool's own subscription or allowance.
* **Adapt a template or existing app:** direct artifact editing happens inside the connected agentic tool, so Buzzy AI generation credits are not required. The agentic tool's subscription or allowance still applies.

If an Adapt workflow explicitly invokes a Buzzy generation tool, the Create-generation requirements apply to that step. For the complete explanation, see [Builder MCP: Create vs Adapt](/working-with-buzzy/buzzy-ai/ai-tokens-and-pricing.md#builder-mcp-create-vs-adapt).

{% hint style="success" %}
**New to Builder MCP? Start with the tutorial.** The hands-on, screenshot-driven [**Tutorial: Build a carer app from a template**](/the-building-blocks/mcp/buzzy-builder-mcp/tutorial-carer-app-from-a-template.md) walks you through a complete build end to end — connect, describe your app, approve each stage, then refine. It's the fastest way to see the whole workflow.
{% endhint %}

## Template adaptation series

Watch each template overview, then use the accompanying brief and checks to build a separate target app:

* [Commerce Insights: a distributor dashboard](/the-building-blocks/mcp/buzzy-builder-mcp/adapt-commerce-insights.md) — agree reporting definitions, shared filters and known expected totals.
* [SitePulse: facilities operations](/the-building-blocks/mcp/buzzy-builder-mcp/adapt-sitepulse.md) — connect sites, signals, alerts and response tasks with visible source freshness.
* [CRM: an agency example](/the-building-blocks/mcp/buzzy-builder-mcp/crm-agency-example.md) — define sales stages, proposals and follow-ups alongside the [CRM customization guide](/the-building-blocks/mcp/buzzy-builder-mcp/customize-crm-template.md).

These are adaptation recipes with demonstration videos, not completed implementations of the proposed target apps. For help choosing a starting pattern, read the [series overview](https://www.buzzy.buzz/post/build-from-templates-builder-mcp/).

## Common Builder Workflows

Builder MCP can start from several different inputs:

* a blank app idea or prompt
* a Buzzy template that is close to your target app
* an existing Buzzy app that needs changes
* an external codebase or legacy app used as a reference
* a Figma file, screenshot set, or design reference

The common thread is staged review. The agent should inspect the starting point, make a plan when the change is broad, update the right Buzzy artifacts, and verify the result in preview or runtime.

### Getting started

Start here once per project. This covers MCP credentials, connecting your agent, bootstrapping the local repo, and the review habits that apply to every Builder MCP workflow.

{% content-ref url="/pages/gOBi8D5yPTX9ofnQJTfT" %}
[Getting started with Builder MCP](/the-building-blocks/mcp/buzzy-builder-mcp/getting-started.md)
{% endcontent-ref %}

### For new apps

Start with an empty repo and a focused app prompt. After setup, the agent creates the app and moves through the brief, flows, data model, theme, blueprint, screens, sample data, and refinement stages.

{% content-ref url="/pages/F2S40cO0BLzrIe5wzsVZ" %}
[For new apps](/the-building-blocks/mcp/buzzy-builder-mcp/for-new-apps.md)
{% endcontent-ref %}

### From a template

Start from an existing Buzzy template when the app shape is close to something Buzzy already provides. The agent should still inspect the template, identify the changes you need, and work through staged edits rather than treating the template as finished.

{% content-ref url="/pages/1t1hGsfoVEqERpYTsk06" %}
[From a template](/the-building-blocks/mcp/buzzy-builder-mcp/from-a-template.md)
{% endcontent-ref %}

To adapt Buzzy's pipeline-first starting point into your own sales system, follow the CRM-specific walkthrough:

{% content-ref url="/pages/LiVFbLRIxBbarVB8qbcK" %}
[Customize the CRM template](/the-building-blocks/mcp/buzzy-builder-mcp/customize-crm-template.md)
{% endcontent-ref %}

Prefer a guided, screenshot-by-screenshot build? Follow the full walkthrough:

{% content-ref url="/pages/V268OApcdnzplyJWVIkn" %}
[Tutorial: Build a carer app from a template](/the-building-blocks/mcp/buzzy-builder-mcp/tutorial-carer-app-from-a-template.md)
{% endcontent-ref %}

### Editing existing apps

Use MCP to help an agent inspect the current app, plan the smallest useful change, apply edits in stages, and verify the result in preview or runtime.

{% content-ref url="/pages/tGFxNoN4FMfFD5MuyUyp" %}
[Editing existing apps](/the-building-blocks/mcp/buzzy-builder-mcp/editing-existing-apps.md)
{% endcontent-ref %}

## Working Style

Good Builder MCP work is collaborative and staged:

* connect the agent to Buzzy MCP and bootstrap the repo first
* review generated artifacts before moving downstream
* use sample data and screenshots when screens exist
* ask for a plan before broad screen or data-model changes
* verify important runtime paths before treating the change as complete

## Accessibility-ready builds

For a WCAG 2.2 AA-targeted build, ask the agent to use Buzzy's accessibility tool sequence rather than relying on a visual preview or one scanner score:

1. inventory the applicable native widgets and functional categories with `get_native_widget_accessibility_matrix`
2. run and repair the persisted-screen preflight with `audit_app_accessibility_structure`
3. verify the real functional journeys
4. persist route-backed browser evidence with `run_app_accessibility_audit`
5. retrieve or compare evidence with `get_app_accessibility_run` and `list_app_accessibility_runs`
6. complete the required manual screen-reader, keyboard, zoom, contrast and content review
7. evaluate the exact selected evidence with `get_app_accessibility_release_gate`

See [Accessibility at Buzzy](/accessibility.md#buzzy-builder-mcp) for the tool capabilities, limitations, full workflow and manual test checklist. Automated evidence alone is not a WCAG conformance claim.
