- Expand `App_ontoolinputpartial_progressiveRendering` example in `app.examples.ts` to show full state management pattern with `toolInputs` and `toolInputsPartial` variables - Add healed JSON caveat to `ontoolinputpartial` JSDoc in `app.ts` - Update `patterns.md` to reference the expanded example and document the streaming preview pattern - Remove empty `lowerPerceivedLatency` placeholder from `patterns.tsx` Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
3.6 KiB
| title |
|---|
| Patterns |
MCP Apps Patterns
This document covers common patterns and recipes for building MCP Apps.
Tools that are private to Apps
Set {@link types!McpUiToolMeta.visibility Tool._meta.ui.visibility} to ["app"] to make tools only callable by Apps (hidden from the model). This is useful for UI-driven actions like updating quantities, toggling settings, or other interactions that shouldn't appear in the model's tool list.
{@includeCode ../src/server/index.examples.ts#registerAppTool_appOnlyVisibility}
[TODO] Authenticated calls from App
- Use tool calls / read resources
- See PDF example to read binaries by chunks to avoid call tool size limitations on platforms like claude.ai
- Pass auth token in
_meta(will be loaded again in the future) + refresh token + store in local storage (see Persist data section below)
{@includeCode ./patterns.tsx#authenticatedCalls}
[TODO] Giving errors back to model
- Before app runs: validate inputs in tool call
- After it runs: use
updateModelContext
{@includeCode ./patterns.tsx#errorsToModel}
Matching host styling (CSS variables, theme, and fonts)
Use the SDK's style helpers to apply host styling, then reference them in your CSS:
- CSS variables — Use
var(--color-background-primary), etc. in your CSS - Theme — Use
[data-theme="dark"]selectors orlight-dark()function for theme-aware styles - Fonts — Use
var(--font-sans)orvar(--font-mono)with fallbacks (e.g.,font-family: var(--font-sans, system-ui, sans-serif))
Vanilla JS:
{@includeCode ./patterns.tsx#hostStylingVanillaJs}
React:
{@includeCode ./patterns.tsx#hostStylingReact}
Entering / Exiting fullscreen
Toggle fullscreen mode by calling {@link app!App.requestDisplayMode requestDisplayMode}:
{@includeCode ../src/app.examples.ts#App_requestDisplayMode_toggle}
Listen for display mode changes via {@link app!App.onhostcontextchanged onhostcontextchanged} to update your UI:
{@includeCode ../src/app.examples.ts#App_onhostcontextchanged_respondToDisplayMode}
[TODO] Persist data (incl. widget state)
- Note: OAI's
window.openai.setWidgetState({modelContent, privateContent, imageIds})has only a partial equivalent in the MCP Apps spec (for now!):App.updateModelContext({content, structuredContent}) - For data persistence / to reload when conversation is reloaded, you must use localStorage / IndexedDb with
hostInfo.toolInfo.idas key returnedCallToolResult._meta.widgetUUID = randomUUID()
{@includeCode ./patterns.tsx#persistData}
Lowering perceived latency
Use {@link app!App.ontoolinputpartial ontoolinputpartial} to receive streaming tool arguments as they arrive, allowing you to show a loading preview before the complete input is available.
{@includeCode ../src/app.examples.ts#App_ontoolinputpartial_progressiveRendering}
Important
Partial arguments are "healed" JSON — the host closes unclosed brackets/braces to produce valid JSON. This means objects may be incomplete (e.g., the last item in an array may be truncated). Don't rely on partial data for critical operations; use it only for preview UI.
[TODO] Supporting both iframe & MCP Apps in same binary
See recipe: https://github.com/modelcontextprotocol/ext-apps/issues/34
{@includeCode ./patterns.tsx#iframeAndMcpApps}
[TODO] Migrating from OpenAI to MCP Apps
See OpenAI -> MCP Apps migration guide.
Also: Managing State (OpenAI)
{@includeCode ./patterns.tsx#migrateFromOpenai}