Migrate from OpenAI App
About
This skill provides step-by-step migration guidance for converting OpenAI Apps SDK applications to the MCP Apps SDK. It helps developers map OpenAI APIs to MCP equivalents and includes API mapping tables for reference. Use it when porting from window.openai, skybridge, or openai/outputTemplate to the standardized @modelcontextprotocol/ext-apps package.
Quick Install
Claude Code
Recommended/plugin add https://github.com/majiayu000/claude-skill-registrygit clone https://github.com/majiayu000/claude-skill-registry.git ~/.claude/skills/Migrate from OpenAI AppCopy and paste this command in Claude Code to install this skill
Documentation
Migrate OpenAI App to MCP
Migrate existing OpenAI Apps SDK applications to the MCP Apps SDK (@modelcontextprotocol/ext-apps). The MCP Apps SDK provides a standardized, open protocol for interactive UIs in conversational clients.
Best Practices
- Use your package manager to add dependencies (e.g.,
npm install,pnpm add,yarn add) instead of manually writing version numbers. This lets the package manager resolve the latest compatible versions. Never specify version numbers from memory. - Preemptively add a final todo item with this exact wording: "Re-read the 'Before Finishing' checklist in this skill and address each checkbox individually, stating what you did for each one, before marking this todo complete."
Getting Reference Code
Clone the SDK repository for complete migration documentation and working examples:
git clone --branch "v$(npm view @modelcontextprotocol/ext-apps version)" --depth 1 https://github.com/modelcontextprotocol/ext-apps.git /tmp/mcp-ext-apps
Migration Reference Guide
Read the migration reference guide with "before/after" mapping tables: /tmp/mcp-ext-apps/docs/migrate_from_openai_apps.md
API Reference (Source Files)
Read JSDoc documentation directly from /tmp/mcp-ext-apps/src/*:
| File | Contents |
|---|---|
src/app.ts | App class, handlers, lifecycle |
src/server/index.ts | registerAppTool, registerAppResource |
src/spec.types.ts | Type definitions |
src/react/useApp.tsx | useApp hook for React apps |
src/react/use*.ts* | Other use* hooks for React apps |
Front-End Framework Examples
See /tmp/mcp-ext-apps/examples/basic-server-{framework}/ for basic SDK usage examples organized by front-end framework:
| Template | Key Files |
|---|---|
basic-server-vanillajs/ | server.ts, src/mcp-app.ts, mcp-app.html |
basic-server-react/ | server.ts, src/mcp-app.tsx (uses useApp hook) |
basic-server-vue/ | server.ts, src/App.vue |
basic-server-svelte/ | server.ts, src/App.svelte |
basic-server-preact/ | server.ts, src/mcp-app.tsx |
basic-server-solid/ | server.ts, src/mcp-app.tsx |
CSP Investigation
MCP Apps HTML is served as an MCP resource, not as a web page, and runs in a sandboxed iframe with no same-origin server. Every origin must be declared in CSP—including the origin serving your JS/CSS bundles (localhost in dev, your CDN in production). Missing origins fail silently.
Before writing any migration code, build the app and investigate all origins it references:
- Build the app using the existing build command
- Search the resulting HTML, CSS, and JS for every origin (not just "external" origins—every network request will need CSP approval)
- For each origin found, trace back to source:
- If it comes from a constant → universal (same in dev and prod)
- If it comes from an env var or conditional → note the mechanism and identify both dev and prod values
- Check for third-party libraries that may make their own requests (analytics, error tracking, etc.)
Document your findings as three lists, and note for each origin whether it's universal, dev-only, or prod-only:
- resourceDomains: origins serving images, fonts, styles, scripts
- connectDomains: origins for API/fetch requests
- frameDomains: origins for nested iframes
If no origins are found, the app may not need custom CSP domains.
CORS Configuration
MCP clients make cross-origin requests. If using Express, app.use(cors()) handles this.
For raw HTTP servers, configure standard CORS and additionally:
- Allow headers:
mcp-session-id,mcp-protocol-version,last-event-id - Expose headers:
mcp-session-id
Key Conceptual Changes
Server-Side
Use registerAppTool() and registerAppResource() helpers instead of raw server.registerTool() / server.registerResource(). These helpers handle the MCP Apps metadata format automatically.
See /tmp/mcp-ext-apps/docs/migrate_from_openai_apps.md for server-side mapping tables.
Client-Side
The fundamental paradigm shift: OpenAI uses a synchronous global object (window.openai.toolInput, window.openai.theme) that's pre-populated before your code runs. MCP Apps uses an App instance with async event handlers.
Key differences:
- Create an
Appinstance and register handlers (ontoolinput,ontoolresult,onhostcontextchanged) before callingconnect(). (Events may fire immediately after connection, so handlers must be registered first.) - Access tool data via handlers:
app.ontoolinputforwindow.openai.toolInput,app.ontoolresultforwindow.openai.toolOutput. - Access host environment (theme, locale, etc.) via
app.getHostContext().
For React apps, the useApp hook manages this lifecycle automatically—see basic-server-react/ for the pattern.
See /tmp/mcp-ext-apps/docs/migrate_from_openai_apps.md for client-side mapping tables.
Features Not Yet Available in MCP Apps
These OpenAI features don't have MCP equivalents yet:
Server-side:
| OpenAI Feature | Status/Workaround |
|---|---|
_meta["openai/toolInvocation/invoking"] / _meta["openai/toolInvocation/invoked"] | Progress indicators not yet available |
_meta["openai/widgetDescription"] | Use app.updateModelContext() for dynamic context |
Client-side:
| OpenAI Feature | Status/Workaround |
|---|---|
window.openai.widgetState / setWidgetState() | Use localStorage or server-side state |
window.openai.uploadFile() / getFileDownloadUrl() | File operations not yet available |
window.openai.requestModal() / requestClose() | Modal management not yet available |
window.openai.view | Not yet available |
Before Finishing
Slow down and carefully follow each item in this checklist:
-
Search for and migrate any remaining server-side OpenAI patterns:
Pattern Indicates "openai/Old metadata keys → _meta.ui.*text/html+skybridgeOld MIME type → RESOURCE_MIME_TYPEconstanttext/html;profile=mcp-appNew MIME type, but prefer RESOURCE_MIME_TYPEconstant_domains"or_domains:snake_case CSP → camelCase ( connect_domains→connectDomains) -
Search for and migrate any remaining client-side OpenAI patterns:
Pattern Indicates window.openai.toolInputOld global → params.argumentsinontoolinputhandlerwindow.openai.toolOutputOld global → params.structuredContentinontoolresultwindow.openaiOld global API → Appinstance methods -
For each origin from your CSP investigation, show where it appears in the
registerAppResource()CSP config. Every origin from the CSP investigation (universal, dev-only, prod-only) must be included in the CSP config—MCP Apps HTML runs in a sandboxed iframe with no same-origin server. If an origin was not included in the CSP config, add it now. -
For each conditional (dev-only, prod-only) origin from your CSP investigation, show the code where the same configuration setting (env var, config file, etc.) controls both the runtime URL and the CSP entry. If the CSP has a hardcoded origin that should be conditional, fix it now—the app must be production-ready.
Testing
Using basic-host
Test the migrated app with the basic-host example:
# Terminal 1: Build and run your server
npm run build && npm run serve
# Terminal 2: Run basic-host (from cloned repo)
cd /tmp/mcp-ext-apps/examples/basic-host
npm install
SERVERS='["http://localhost:3001/mcp"]' npm run start
# Open http://localhost:8080
Verify Runtime Behavior
Once the app loads in basic-host, confirm:
- App loads without console errors
ontoolinputhandler fires with tool argumentsontoolresulthandler fires with tool result
GitHub Repository
Related Skills
content-collections
MetaThis skill provides a production-tested setup for Content Collections, a TypeScript-first tool that transforms Markdown/MDX files into type-safe data collections with Zod validation. Use it when building blogs, documentation sites, or content-heavy Vite + React applications to ensure type safety and automatic content validation. It covers everything from Vite plugin configuration and MDX compilation to deployment optimization and schema validation.
creating-opencode-plugins
MetaThis skill provides the structure and API specifications for creating OpenCode plugins that hook into 25+ event types like commands, files, and LSP operations. It offers implementation patterns for JavaScript/TypeScript modules that intercept and extend the AI assistant's lifecycle. Use it when you need to build event-driven plugins for monitoring, custom handling, or extending OpenCode's capabilities.
sglang
MetaSGLang is a high-performance LLM serving framework that specializes in fast, structured generation for JSON, regex, and agentic workflows using its RadixAttention prefix caching. It delivers significantly faster inference, especially for tasks with repeated prefixes, making it ideal for complex, structured outputs and multi-turn conversations. Choose SGLang over alternatives like vLLM when you need constrained decoding or are building applications with extensive prefix sharing.
evaluating-llms-harness
TestingThis Claude Skill runs the lm-evaluation-harness to benchmark LLMs across 60+ standardized academic tasks like MMLU and GSM8K. It's designed for developers to compare model quality, track training progress, or report academic results. The tool supports various backends including HuggingFace and vLLM models.
