/**
 * Streamable-HTTP MCP server for one design-canvas document — the live
 * agent→canvas channel. JSON-RPC 2.0 over POST, stateless per request
 * (Workers-compatible: no session table, no SSE).
 *
 * Trust boundary: the product authenticates the request (capability token,
 * workspace RBAC) BEFORE constructing the scoped {@link SceneStore} and
 * calling this handler — the handler trusts its store completely. Tool
 * execution failures (argument shape, validation, stale rev) become `isError`
 * tool results carrying the thrown message verbatim so the model can read WHY
 * and retry.
 *
 * The JSON-RPC envelope (initialize/ping/tools/list/tools/call, -32601) lives
 * in {@link createMcpToolHandler}; this module wires the canvas tool list +
 * the caller-supplied id-minting function.
 */
import type { SceneStore } from './store';
/** Describe design canvas MCP server information including name and version */
export interface DesignCanvasMcpServerInfo {
    name: string;
    version: string;
}
/** Define options for creating a design canvas MCP handler including store, ID minting, and optional server info */
export interface CreateDesignCanvasMcpHandlerOptions {
    /** Already scoped + authorized for one (workspace, document, actor). */
    store: SceneStore;
    /** Id-minting function threaded into every mutating tool. Must return
     *  a string unique within the document scope — use crypto.randomUUID()
     *  in production; a deterministic counter in tests. */
    mintId: () => string;
    serverInfo?: DesignCanvasMcpServerInfo;
}
/** Create a request handler for the design canvas MCP tool with specified options */
export declare function createDesignCanvasMcpHandler(opts: CreateDesignCanvasMcpHandlerOptions): (request: Request) => Promise<Response>;