UNPKG

ai

Version:

AI SDK by Vercel - build apps like ChatGPT, Claude, Gemini, and more with a single interface for any model using the Vercel AI Gateway or go direct to OpenAI, Anthropic, Google, or any other model provider.

184 lines (150 loc) 4.01 kB
--- title: MCP Apps description: API reference for MCP Apps helpers in @ai-sdk/mcp. --- # MCP Apps The MCP Apps helpers in `@ai-sdk/mcp` help an MCP host advertise UI support, keep model-visible and app-visible tools separate, and read `ui://` HTML resources for rendering. ## Import <Snippet text={`import { MCP_APP_MIME_TYPE, mcpAppClientCapabilities, readMCPAppResource, splitMCPAppTools, } from "@ai-sdk/mcp"`} prompt={false} /> ## `MCP_APP_MIME_TYPE` The MIME type for HTML resources that should be rendered as MCP Apps. ```ts const MCP_APP_MIME_TYPE = 'text/html;profile=mcp-app'; ``` ## `mcpAppClientCapabilities` Client capabilities to pass to [`createMCPClient`](/docs/reference/ai-sdk-core/create-mcp-client) when your host supports MCP Apps. ```ts import { createMCPClient, mcpAppClientCapabilities } from '@ai-sdk/mcp'; const client = await createMCPClient({ transport: { type: 'http', url: 'https://example.com/mcp', }, capabilities: mcpAppClientCapabilities, }); ``` The advertised capability is: ```json { "extensions": { "io.modelcontextprotocol/ui": { "mimeTypes": ["text/html;profile=mcp-app"] } } } ``` ## `splitMCPAppTools()` Splits MCP tool definitions into model-visible tools and app-visible tools. Tools without MCP Apps visibility metadata remain model-visible. Tools whose `_meta.ui.visibility` includes `"app"` are returned in `appVisible`. ```ts const definitions = await client.listTools(); const { modelVisible, appVisible } = splitMCPAppTools(definitions); const tools = client.toolsFromDefinitions(modelVisible); ``` ### Parameters <PropertiesTable content={[ { name: 'definitions', type: 'ListToolsResult', description: 'The tool definitions returned by `client.listTools()`.', }, ]} /> ### Returns <PropertiesTable content={[ { name: 'modelVisible', type: 'ListToolsResult', description: 'Tool definitions that can be exposed to the language model.', }, { name: 'appVisible', type: 'ListToolsResult', description: 'Tool definitions that can be called by an MCP App through the host bridge.', }, ]} /> ## `readMCPAppResource()` Reads a `ui://` resource from an MCP server and normalizes it into HTML plus rendering metadata. ```ts const resource = await readMCPAppResource({ client, uri: 'ui://example/dashboard', }); ``` The helper validates that the URI starts with `ui://`, requires the `text/html;profile=mcp-app` MIME type, and supports resource contents returned as either text or base64 blob data. ### Parameters <PropertiesTable content={[ { name: 'client', type: "Pick<MCPClient, 'readResource'>", description: 'The MCP client used to read the resource.', }, { name: 'uri', type: 'string', description: 'The `ui://` resource URI to read.', }, { name: 'options', type: 'RequestOptions', isOptional: true, description: 'Optional request options, such as an abort signal or timeout.', }, ]} /> ### Returns Returns a `Promise<MCPAppResource>`. <PropertiesTable content={[ { name: 'uri', type: 'string', description: 'The `ui://` resource URI.', }, { name: 'mimeType', type: "'text/html;profile=mcp-app'", description: 'The MCP Apps HTML MIME type.', }, { name: 'html', type: 'string', description: 'The app HTML to render in a sandboxed iframe.', }, { name: 'meta', type: 'MCPAppResourceMeta', isOptional: true, description: 'Rendering metadata from resource `_meta.ui`, such as CSP, permissions, and `prefersBorder`.', }, ]} /> ## See Also <ExampleLinks examples={[ { title: 'MCP Apps guide', link: '/docs/ai-sdk-core/mcp-apps', }, { title: 'createMCPClient', link: '/docs/reference/ai-sdk-core/create-mcp-client', }, ]} />