@yourgpt/llm-sdk 2.5.1-beta.0 → 2.5.1-beta.1

package/dist/index.d.mts

@@ -1,13 +1,13 @@
-import { G as GenerateTextParams, a as GenerateTextResult, S as StreamTextParams, b as StreamTextResult, T as ToolContext, c as Tool, A as ActionDefinition, d as ToolDefinition, e as ToolProfile, K as KnowledgeBaseConfig, W as WebSearchConfig, f as StorageAdapter, R as ResponseFormat, D as DoneEventMessage, g as StreamEvent, h as ToolCallInfo, i as TokenUsageRaw, P as ProviderToolRuntimeOptions, M as Message, j as ToolResponse, k as StorageMessage } from './types-BkQCSiIt.mjs';
-export { a4 as AnthropicProviderToolOptions, a1 as AnthropicToolSelectionHints, p as AssistantMessage, a5 as Citation, C as CoreMessage, V as DEFAULT_CAPABILITIES, m as DoGenerateParams, n as DoGenerateResult, H as ErrorChunk, F as FilePart, E as FinishChunk, N as FinishReason, v as GenerateStep, I as ImagePart, X as LLMConfig, L as LanguageModel, l as ModelCapabilities, a3 as OpenAIProviderToolOptions, a0 as OpenAIToolSelectionHints, O as ResponseOptions, Q as StorageFile, x as StreamChunk, w as StreamPart, o as SystemMessage, y as TextDeltaChunk, s as TextPart, J as TokenUsage, t as ToolCall, z as ToolCallChunk, $ as ToolExecution, Y as ToolLocation, q as ToolMessage, a2 as ToolNativeProviderHints, u as ToolResult, B as ToolResultChunk, Z as UnifiedToolCall, _ as UnifiedToolResult, r as UserContentPart, U as UserMessage } from './types-BkQCSiIt.mjs';
+import { G as GenerateTextParams, a as GenerateTextResult, S as StreamTextParams, b as StreamTextResult, T as ToolContext, c as Tool, A as ActionDefinition, d as ToolDefinition, e as ToolProfile, K as KnowledgeBaseConfig, W as WebSearchConfig, f as StorageAdapter, R as ResponseFormat, M as McpServerConfig, g as ReasoningEffort, D as DoneEventMessage, h as StreamEvent, i as ToolCallInfo, j as TokenUsageRaw, P as ProviderToolRuntimeOptions, k as Message, l as ToolResponse, m as StorageMessage } from './types-D774b0dg.mjs';
+export { a6 as AnthropicProviderToolOptions, a3 as AnthropicToolSelectionHints, r as AssistantMessage, a7 as Citation, C as CoreMessage, Y as DEFAULT_CAPABILITIES, o as DoGenerateParams, p as DoGenerateResult, N as ErrorChunk, F as FilePart, J as FinishChunk, Q as FinishReason, x as GenerateStep, I as ImagePart, Z as LLMConfig, L as LanguageModel, n as ModelCapabilities, a5 as OpenAIProviderToolOptions, a2 as OpenAIToolSelectionHints, V as ResponseOptions, X as StorageFile, z as StreamChunk, y as StreamPart, q as SystemMessage, B as TextDeltaChunk, u as TextPart, O as TokenUsage, v as ToolCall, E as ToolCallChunk, a1 as ToolExecution, _ as ToolLocation, s as ToolMessage, a4 as ToolNativeProviderHints, w as ToolResult, H as ToolResultChunk, $ as UnifiedToolCall, a0 as UnifiedToolResult, t as UserContentPart, U as UserMessage } from './types-D774b0dg.mjs';
 import { z } from 'zod';
-import { A as AIProvider } from './types-BSSiJW2o.mjs';
-export { a as AnthropicProviderConfig, f as AnthropicTool, h as AnthropicToolResult, g as AnthropicToolUse, b as AzureProviderConfig, B as BaseProviderConfig, m as GeminiFunctionCall, l as GeminiFunctionDeclaration, n as GeminiFunctionResponse, G as GoogleProviderConfig, d as OllamaModelOptions, c as OllamaProviderConfig, O as OpenAIProviderConfig, i as OpenAITool, j as OpenAIToolCall, k as OpenAIToolResult, P as ProviderCapabilities, e as ProviderFormatter, X as XAIProviderConfig } from './types-BSSiJW2o.mjs';
-import { L as LLMAdapter } from './base-tNgbBaSo.mjs';
-export { A as AdapterFactory, C as ChatCompletionRequest } from './base-tNgbBaSo.mjs';
+import { A as AIProvider } from './types-BvkiJ1dd.mjs';
+export { a as AnthropicProviderConfig, f as AnthropicTool, h as AnthropicToolResult, g as AnthropicToolUse, b as AzureProviderConfig, B as BaseProviderConfig, m as GeminiFunctionCall, l as GeminiFunctionDeclaration, n as GeminiFunctionResponse, G as GoogleProviderConfig, d as OllamaModelOptions, c as OllamaProviderConfig, O as OpenAIProviderConfig, i as OpenAITool, j as OpenAIToolCall, k as OpenAIToolResult, P as ProviderCapabilities, e as ProviderFormatter, X as XAIProviderConfig } from './types-BvkiJ1dd.mjs';
+import { L as LLMAdapter } from './base-BYQKp9TW.mjs';
+export { A as AdapterFactory, C as ChatCompletionRequest } from './base-BYQKp9TW.mjs';
 import * as hono from 'hono';
 import { Hono } from 'hono';
-export { F as FallbackChainConfig, c as FallbackFailure, d as FallbackInfo, b as RetryBackoff, e as RetryInfo, R as RoutingStore, a as RoutingStrategy } from './types-B6dhnguR.mjs';
+export { F as FallbackChainConfig, c as FallbackFailure, d as FallbackInfo, b as RetryBackoff, e as RetryInfo, R as RoutingStore, a as RoutingStrategy } from './types-mwMhCwOq.mjs';
 
 /**
  * generateText - Generate text using a language model
@@ -389,6 +389,8 @@ interface ChatRequest {
         temperature?: number;
         maxTokens?: number;
         responseFormat?: ResponseFormat;
+        mcpServers?: McpServerConfig[];
+        reasoningEffort?: ReasoningEffort;
     };
     /** System prompt override */
     systemPrompt?: string;
@@ -798,6 +800,11 @@ declare function createStreamResult(generator: AsyncGenerator<StreamEvent>): Str
  * ```
  */
 
+interface GenerateUsage {
+    promptTokens: number;
+    completionTokens: number;
+    totalTokens: number;
+}
 interface GenerateResultData {
     text: string;
     messages: DoneEventMessage[];
@@ -811,6 +818,7 @@ interface GenerateResultData {
         result: unknown;
     }>;
     requiresAction: boolean;
+    usage?: GenerateUsage;
     error?: {
         message: string;
         code?: string;
@@ -864,6 +872,12 @@ declare class GenerateResult {
      * Whether client action is required (e.g., tool approval)
      */
     get requiresAction(): boolean;
+    /**
+     * Token usage for the call — present when the underlying provider emitted
+     * a usage payload on the `done` event. May be undefined for streaming
+     * responses that don't yield usage (e.g. mid-stream errors).
+     */
+    get usage(): GenerateUsage | undefined;
     /**
      * Error if generation failed
      */
@@ -1123,6 +1137,44 @@ declare class Runtime {
      * ```
      */
     generate(request: ChatRequest, options?: GenerateOptions): Promise<GenerateResult>;
+    /**
+     * One-shot non-streaming call bundling MCP servers, reasoning effort, and
+     * structured output. Thin ergonomic wrapper around `generate()` — uses the
+     * same adapter chain and the same translators, just expressed as a single
+     * prompt-in / text-out call.
+     *
+     * @example
+     * ```ts
+     * const result = await runtime.response({
+     *   prompt: "Extract FAQs from this conversation.",
+     *   mcpServers: [{ label: "kb", url: "https://kb.example.com/sse", headers: { Authorization: token } }],
+     *   reasoningEffort: "high",
+     *   responseFormat: { type: "json_schema", json_schema: { name: "faqs", schema } },
+     * });
+     * const data = JSON.parse(result.text);
+     * ```
+     */
+    response(request: {
+        prompt: string;
+        systemPrompt?: string;
+        mcpServers?: McpServerConfig[];
+        reasoningEffort?: ReasoningEffort;
+        responseFormat?: ResponseFormat;
+        maxTokens?: number;
+        temperature?: number;
+    }): Promise<{
+        text: string;
+        toolCalls: Array<{
+            id: string;
+            name: string;
+            args: Record<string, unknown>;
+        }>;
+        usage?: {
+            promptTokens: number;
+            completionTokens: number;
+            totalTokens: number;
+        };
+    }>;
     /**
      * Create Express-compatible handler middleware
      *
@@ -1537,4 +1589,4 @@ declare function generateThreadId(): string;
  */
 declare function generateToolCallId(): string;
 
-export { AIProvider, ActionDefinition, type ActionRequest, type AgentLoopOptions, type ChatRequest, type CollectedResult, type CopilotChatResponse, DEFAULT_MAX_ITERATIONS, DoneEventMessage, GenerateResult, type GenerateResultData, GenerateTextParams, GenerateTextResult, LLMAdapter, Message, ProviderToolRuntimeOptions, type RequestContext, Runtime, type RuntimeConfig, StorageAdapter, StorageMessage, StreamEvent, StreamResult, type StreamResultOptions, StreamTextParams, StreamTextResult, TokenUsageRaw, Tool, ToolCallInfo, ToolContext, ToolDefinition, ToolProfile, ToolResponse, type ToolSearchMatch, WebSearchConfig, buildProviderToolOptions, createEventStream, createExpressHandler, createExpressMiddleware, createHonoApp, createNextHandler, createNodeHandler, createRuntime, createSSEHeaders, createSSEResponse, createStreamResult, createTextStreamHeaders, createTextStreamResponse, extractInputMessages, formatSSEData, formatToolsForAnthropic, formatToolsForGoogle, formatToolsForOpenAI, generateMessageId, generateText, generateThreadId, generateToolCallId, mapOutputMessages, pipeSSEToResponse, pipeTextToResponse, runAgentLoop, searchTools, selectTools, shouldExposeToolSearch, streamText, tool };
+export { AIProvider, ActionDefinition, type ActionRequest, type AgentLoopOptions, type ChatRequest, type CollectedResult, type CopilotChatResponse, DEFAULT_MAX_ITERATIONS, DoneEventMessage, GenerateResult, type GenerateResultData, GenerateTextParams, GenerateTextResult, LLMAdapter, McpServerConfig, Message, ProviderToolRuntimeOptions, ReasoningEffort, type RequestContext, ResponseFormat, Runtime, type RuntimeConfig, StorageAdapter, StorageMessage, StreamEvent, StreamResult, type StreamResultOptions, StreamTextParams, StreamTextResult, TokenUsageRaw, Tool, ToolCallInfo, ToolContext, ToolDefinition, ToolProfile, ToolResponse, type ToolSearchMatch, WebSearchConfig, buildProviderToolOptions, createEventStream, createExpressHandler, createExpressMiddleware, createHonoApp, createNextHandler, createNodeHandler, createRuntime, createSSEHeaders, createSSEResponse, createStreamResult, createTextStreamHeaders, createTextStreamResponse, extractInputMessages, formatSSEData, formatToolsForAnthropic, formatToolsForGoogle, formatToolsForOpenAI, generateMessageId, generateText, generateThreadId, generateToolCallId, mapOutputMessages, pipeSSEToResponse, pipeTextToResponse, runAgentLoop, searchTools, selectTools, shouldExposeToolSearch, streamText, tool };

package/dist/index.d.ts

@@ -1,13 +1,13 @@
-import { G as GenerateTextParams, a as GenerateTextResult, S as StreamTextParams, b as StreamTextResult, T as ToolContext, c as Tool, A as ActionDefinition, d as ToolDefinition, e as ToolProfile, K as KnowledgeBaseConfig, W as WebSearchConfig, f as StorageAdapter, R as ResponseFormat, D as DoneEventMessage, g as StreamEvent, h as ToolCallInfo, i as TokenUsageRaw, P as ProviderToolRuntimeOptions, M as Message, j as ToolResponse, k as StorageMessage } from './types-BkQCSiIt.js';
-export { a4 as AnthropicProviderToolOptions, a1 as AnthropicToolSelectionHints, p as AssistantMessage, a5 as Citation, C as CoreMessage, V as DEFAULT_CAPABILITIES, m as DoGenerateParams, n as DoGenerateResult, H as ErrorChunk, F as FilePart, E as FinishChunk, N as FinishReason, v as GenerateStep, I as ImagePart, X as LLMConfig, L as LanguageModel, l as ModelCapabilities, a3 as OpenAIProviderToolOptions, a0 as OpenAIToolSelectionHints, O as ResponseOptions, Q as StorageFile, x as StreamChunk, w as StreamPart, o as SystemMessage, y as TextDeltaChunk, s as TextPart, J as TokenUsage, t as ToolCall, z as ToolCallChunk, $ as ToolExecution, Y as ToolLocation, q as ToolMessage, a2 as ToolNativeProviderHints, u as ToolResult, B as ToolResultChunk, Z as UnifiedToolCall, _ as UnifiedToolResult, r as UserContentPart, U as UserMessage } from './types-BkQCSiIt.js';
+import { G as GenerateTextParams, a as GenerateTextResult, S as StreamTextParams, b as StreamTextResult, T as ToolContext, c as Tool, A as ActionDefinition, d as ToolDefinition, e as ToolProfile, K as KnowledgeBaseConfig, W as WebSearchConfig, f as StorageAdapter, R as ResponseFormat, M as McpServerConfig, g as ReasoningEffort, D as DoneEventMessage, h as StreamEvent, i as ToolCallInfo, j as TokenUsageRaw, P as ProviderToolRuntimeOptions, k as Message, l as ToolResponse, m as StorageMessage } from './types-D774b0dg.js';
+export { a6 as AnthropicProviderToolOptions, a3 as AnthropicToolSelectionHints, r as AssistantMessage, a7 as Citation, C as CoreMessage, Y as DEFAULT_CAPABILITIES, o as DoGenerateParams, p as DoGenerateResult, N as ErrorChunk, F as FilePart, J as FinishChunk, Q as FinishReason, x as GenerateStep, I as ImagePart, Z as LLMConfig, L as LanguageModel, n as ModelCapabilities, a5 as OpenAIProviderToolOptions, a2 as OpenAIToolSelectionHints, V as ResponseOptions, X as StorageFile, z as StreamChunk, y as StreamPart, q as SystemMessage, B as TextDeltaChunk, u as TextPart, O as TokenUsage, v as ToolCall, E as ToolCallChunk, a1 as ToolExecution, _ as ToolLocation, s as ToolMessage, a4 as ToolNativeProviderHints, w as ToolResult, H as ToolResultChunk, $ as UnifiedToolCall, a0 as UnifiedToolResult, t as UserContentPart, U as UserMessage } from './types-D774b0dg.js';
 import { z } from 'zod';
-import { A as AIProvider } from './types-BQ31QIsA.js';
-export { a as AnthropicProviderConfig, f as AnthropicTool, h as AnthropicToolResult, g as AnthropicToolUse, b as AzureProviderConfig, B as BaseProviderConfig, m as GeminiFunctionCall, l as GeminiFunctionDeclaration, n as GeminiFunctionResponse, G as GoogleProviderConfig, d as OllamaModelOptions, c as OllamaProviderConfig, O as OpenAIProviderConfig, i as OpenAITool, j as OpenAIToolCall, k as OpenAIToolResult, P as ProviderCapabilities, e as ProviderFormatter, X as XAIProviderConfig } from './types-BQ31QIsA.js';
-import { L as LLMAdapter } from './base-C58Dsr9p.js';
-export { A as AdapterFactory, C as ChatCompletionRequest } from './base-C58Dsr9p.js';
+import { A as AIProvider } from './types-TMilS-Dz.js';
+export { a as AnthropicProviderConfig, f as AnthropicTool, h as AnthropicToolResult, g as AnthropicToolUse, b as AzureProviderConfig, B as BaseProviderConfig, m as GeminiFunctionCall, l as GeminiFunctionDeclaration, n as GeminiFunctionResponse, G as GoogleProviderConfig, d as OllamaModelOptions, c as OllamaProviderConfig, O as OpenAIProviderConfig, i as OpenAITool, j as OpenAIToolCall, k as OpenAIToolResult, P as ProviderCapabilities, e as ProviderFormatter, X as XAIProviderConfig } from './types-TMilS-Dz.js';
+import { L as LLMAdapter } from './base-Cxq3ni0t.js';
+export { A as AdapterFactory, C as ChatCompletionRequest } from './base-Cxq3ni0t.js';
 import * as hono from 'hono';
 import { Hono } from 'hono';
-export { F as FallbackChainConfig, c as FallbackFailure, d as FallbackInfo, b as RetryBackoff, e as RetryInfo, R as RoutingStore, a as RoutingStrategy } from './types-CCxPmkmK.js';
+export { F as FallbackChainConfig, c as FallbackFailure, d as FallbackInfo, b as RetryBackoff, e as RetryInfo, R as RoutingStore, a as RoutingStrategy } from './types-ChORafYS.js';
 
 /**
  * generateText - Generate text using a language model
@@ -389,6 +389,8 @@ interface ChatRequest {
         temperature?: number;
         maxTokens?: number;
         responseFormat?: ResponseFormat;
+        mcpServers?: McpServerConfig[];
+        reasoningEffort?: ReasoningEffort;
     };
     /** System prompt override */
     systemPrompt?: string;
@@ -798,6 +800,11 @@ declare function createStreamResult(generator: AsyncGenerator<StreamEvent>): Str
  * ```
  */
 
+interface GenerateUsage {
+    promptTokens: number;
+    completionTokens: number;
+    totalTokens: number;
+}
 interface GenerateResultData {
     text: string;
     messages: DoneEventMessage[];
@@ -811,6 +818,7 @@ interface GenerateResultData {
         result: unknown;
     }>;
     requiresAction: boolean;
+    usage?: GenerateUsage;
     error?: {
         message: string;
         code?: string;
@@ -864,6 +872,12 @@ declare class GenerateResult {
      * Whether client action is required (e.g., tool approval)
      */
     get requiresAction(): boolean;
+    /**
+     * Token usage for the call — present when the underlying provider emitted
+     * a usage payload on the `done` event. May be undefined for streaming
+     * responses that don't yield usage (e.g. mid-stream errors).
+     */
+    get usage(): GenerateUsage | undefined;
     /**
      * Error if generation failed
      */
@@ -1123,6 +1137,44 @@ declare class Runtime {
      * ```
      */
     generate(request: ChatRequest, options?: GenerateOptions): Promise<GenerateResult>;
+    /**
+     * One-shot non-streaming call bundling MCP servers, reasoning effort, and
+     * structured output. Thin ergonomic wrapper around `generate()` — uses the
+     * same adapter chain and the same translators, just expressed as a single
+     * prompt-in / text-out call.
+     *
+     * @example
+     * ```ts
+     * const result = await runtime.response({
+     *   prompt: "Extract FAQs from this conversation.",
+     *   mcpServers: [{ label: "kb", url: "https://kb.example.com/sse", headers: { Authorization: token } }],
+     *   reasoningEffort: "high",
+     *   responseFormat: { type: "json_schema", json_schema: { name: "faqs", schema } },
+     * });
+     * const data = JSON.parse(result.text);
+     * ```
+     */
+    response(request: {
+        prompt: string;
+        systemPrompt?: string;
+        mcpServers?: McpServerConfig[];
+        reasoningEffort?: ReasoningEffort;
+        responseFormat?: ResponseFormat;
+        maxTokens?: number;
+        temperature?: number;
+    }): Promise<{
+        text: string;
+        toolCalls: Array<{
+            id: string;
+            name: string;
+            args: Record<string, unknown>;
+        }>;
+        usage?: {
+            promptTokens: number;
+            completionTokens: number;
+            totalTokens: number;
+        };
+    }>;
     /**
      * Create Express-compatible handler middleware
      *
@@ -1537,4 +1589,4 @@ declare function generateThreadId(): string;
  */
 declare function generateToolCallId(): string;
 
-export { AIProvider, ActionDefinition, type ActionRequest, type AgentLoopOptions, type ChatRequest, type CollectedResult, type CopilotChatResponse, DEFAULT_MAX_ITERATIONS, DoneEventMessage, GenerateResult, type GenerateResultData, GenerateTextParams, GenerateTextResult, LLMAdapter, Message, ProviderToolRuntimeOptions, type RequestContext, Runtime, type RuntimeConfig, StorageAdapter, StorageMessage, StreamEvent, StreamResult, type StreamResultOptions, StreamTextParams, StreamTextResult, TokenUsageRaw, Tool, ToolCallInfo, ToolContext, ToolDefinition, ToolProfile, ToolResponse, type ToolSearchMatch, WebSearchConfig, buildProviderToolOptions, createEventStream, createExpressHandler, createExpressMiddleware, createHonoApp, createNextHandler, createNodeHandler, createRuntime, createSSEHeaders, createSSEResponse, createStreamResult, createTextStreamHeaders, createTextStreamResponse, extractInputMessages, formatSSEData, formatToolsForAnthropic, formatToolsForGoogle, formatToolsForOpenAI, generateMessageId, generateText, generateThreadId, generateToolCallId, mapOutputMessages, pipeSSEToResponse, pipeTextToResponse, runAgentLoop, searchTools, selectTools, shouldExposeToolSearch, streamText, tool };
+export { AIProvider, ActionDefinition, type ActionRequest, type AgentLoopOptions, type ChatRequest, type CollectedResult, type CopilotChatResponse, DEFAULT_MAX_ITERATIONS, DoneEventMessage, GenerateResult, type GenerateResultData, GenerateTextParams, GenerateTextResult, LLMAdapter, McpServerConfig, Message, ProviderToolRuntimeOptions, ReasoningEffort, type RequestContext, ResponseFormat, Runtime, type RuntimeConfig, StorageAdapter, StorageMessage, StreamEvent, StreamResult, type StreamResultOptions, StreamTextParams, StreamTextResult, TokenUsageRaw, Tool, ToolCallInfo, ToolContext, ToolDefinition, ToolProfile, ToolResponse, type ToolSearchMatch, WebSearchConfig, buildProviderToolOptions, createEventStream, createExpressHandler, createExpressMiddleware, createHonoApp, createNextHandler, createNodeHandler, createRuntime, createSSEHeaders, createSSEResponse, createStreamResult, createTextStreamHeaders, createTextStreamResponse, extractInputMessages, formatSSEData, formatToolsForAnthropic, formatToolsForGoogle, formatToolsForOpenAI, generateMessageId, generateText, generateThreadId, generateToolCallId, mapOutputMessages, pipeSSEToResponse, pipeTextToResponse, runAgentLoop, searchTools, selectTools, shouldExposeToolSearch, streamText, tool };

package/dist/index.js

@@ -155,6 +155,8 @@ async function generateText(params) {
       temperature: params.temperature,
       maxTokens: params.maxTokens,
       responseFormat: params.responseFormat,
+      mcpServers: params.mcpServers,
+      reasoningEffort: params.reasoningEffort,
       signal
     });
     const stepToolResults = [];
@@ -303,6 +305,8 @@ async function streamText(params) {
           temperature: params.temperature,
           maxTokens: params.maxTokens,
           responseFormat: params.responseFormat,
+          mcpServers: params.mcpServers,
+          reasoningEffort: params.reasoningEffort,
           signal
         })) {
           switch (chunk.type) {
@@ -1171,6 +1175,14 @@ var GenerateResult = class {
   get requiresAction() {
     return this.data.requiresAction;
   }
+  /**
+   * Token usage for the call — present when the underlying provider emitted
+   * a usage payload on the `done` event. May be undefined for streaming
+   * responses that don't yield usage (e.g. mid-stream errors).
+   */
+  get usage() {
+    return this.data.usage;
+  }
   /**
    * Error if generation failed
    */
@@ -2957,6 +2969,7 @@ var Runtime = class {
     const toolResults = [];
     let messages = [];
     let requiresAction = false;
+    let usage;
     let error;
     try {
       for await (const event of generator) {
@@ -2987,6 +3000,13 @@ var Runtime = class {
           case "done":
             messages = event.messages || [];
             requiresAction = event.requiresAction || false;
+            if (event.usage) {
+              usage = {
+                promptTokens: event.usage.prompt_tokens,
+                completionTokens: event.usage.completion_tokens,
+                totalTokens: event.usage.total_tokens ?? event.usage.prompt_tokens + event.usage.completion_tokens
+              };
+            }
             break;
           case "error":
             error = { message: event.message, code: event.code };
@@ -3018,9 +3038,48 @@ var Runtime = class {
       toolCalls,
       toolResults,
       requiresAction,
+      usage,
       error
     });
   }
+  /**
+   * One-shot non-streaming call bundling MCP servers, reasoning effort, and
+   * structured output. Thin ergonomic wrapper around `generate()` — uses the
+   * same adapter chain and the same translators, just expressed as a single
+   * prompt-in / text-out call.
+   *
+   * @example
+   * ```ts
+   * const result = await runtime.response({
+   *   prompt: "Extract FAQs from this conversation.",
+   *   mcpServers: [{ label: "kb", url: "https://kb.example.com/sse", headers: { Authorization: token } }],
+   *   reasoningEffort: "high",
+   *   responseFormat: { type: "json_schema", json_schema: { name: "faqs", schema } },
+   * });
+   * const data = JSON.parse(result.text);
+   * ```
+   */
+  async response(request) {
+    const result = await this.generate({
+      messages: [{ role: "user", content: request.prompt }],
+      systemPrompt: request.systemPrompt,
+      config: {
+        temperature: request.temperature,
+        maxTokens: request.maxTokens,
+        responseFormat: request.responseFormat,
+        mcpServers: request.mcpServers,
+        reasoningEffort: request.reasoningEffort
+      }
+    });
+    if (result.error) {
+      throw new Error(`[llm-sdk] response() failed: ${result.error.message}`);
+    }
+    return {
+      text: result.text,
+      toolCalls: result.toolCalls,
+      usage: result.usage
+    };
+  }
   /**
    * Create Express-compatible handler middleware
    *

package/dist/index.mjs

@@ -153,6 +153,8 @@ async function generateText(params) {
       temperature: params.temperature,
       maxTokens: params.maxTokens,
       responseFormat: params.responseFormat,
+      mcpServers: params.mcpServers,
+      reasoningEffort: params.reasoningEffort,
       signal
     });
     const stepToolResults = [];
@@ -301,6 +303,8 @@ async function streamText(params) {
           temperature: params.temperature,
           maxTokens: params.maxTokens,
           responseFormat: params.responseFormat,
+          mcpServers: params.mcpServers,
+          reasoningEffort: params.reasoningEffort,
           signal
         })) {
           switch (chunk.type) {
@@ -1169,6 +1173,14 @@ var GenerateResult = class {
   get requiresAction() {
     return this.data.requiresAction;
   }
+  /**
+   * Token usage for the call — present when the underlying provider emitted
+   * a usage payload on the `done` event. May be undefined for streaming
+   * responses that don't yield usage (e.g. mid-stream errors).
+   */
+  get usage() {
+    return this.data.usage;
+  }
   /**
    * Error if generation failed
    */
@@ -2955,6 +2967,7 @@ var Runtime = class {
     const toolResults = [];
     let messages = [];
     let requiresAction = false;
+    let usage;
     let error;
     try {
       for await (const event of generator) {
@@ -2985,6 +2998,13 @@ var Runtime = class {
           case "done":
             messages = event.messages || [];
             requiresAction = event.requiresAction || false;
+            if (event.usage) {
+              usage = {
+                promptTokens: event.usage.prompt_tokens,
+                completionTokens: event.usage.completion_tokens,
+                totalTokens: event.usage.total_tokens ?? event.usage.prompt_tokens + event.usage.completion_tokens
+              };
+            }
             break;
           case "error":
             error = { message: event.message, code: event.code };
@@ -3016,9 +3036,48 @@ var Runtime = class {
       toolCalls,
       toolResults,
       requiresAction,
+      usage,
       error
     });
   }
+  /**
+   * One-shot non-streaming call bundling MCP servers, reasoning effort, and
+   * structured output. Thin ergonomic wrapper around `generate()` — uses the
+   * same adapter chain and the same translators, just expressed as a single
+   * prompt-in / text-out call.
+   *
+   * @example
+   * ```ts
+   * const result = await runtime.response({
+   *   prompt: "Extract FAQs from this conversation.",
+   *   mcpServers: [{ label: "kb", url: "https://kb.example.com/sse", headers: { Authorization: token } }],
+   *   reasoningEffort: "high",
+   *   responseFormat: { type: "json_schema", json_schema: { name: "faqs", schema } },
+   * });
+   * const data = JSON.parse(result.text);
+   * ```
+   */
+  async response(request) {
+    const result = await this.generate({
+      messages: [{ role: "user", content: request.prompt }],
+      systemPrompt: request.systemPrompt,
+      config: {
+        temperature: request.temperature,
+        maxTokens: request.maxTokens,
+        responseFormat: request.responseFormat,
+        mcpServers: request.mcpServers,
+        reasoningEffort: request.reasoningEffort
+      }
+    });
+    if (result.error) {
+      throw new Error(`[llm-sdk] response() failed: ${result.error.message}`);
+    }
+    return {
+      text: result.text,
+      toolCalls: result.toolCalls,
+      usage: result.usage
+    };
+  }
   /**
    * Create Express-compatible handler middleware
    *

package/dist/providers/anthropic/index.d.mts

@@ -1,7 +1,7 @@
-import { L as LanguageModel } from '../../types-BkQCSiIt.mjs';
-import { a as AnthropicProviderConfig, A as AIProvider } from '../../types-BSSiJW2o.mjs';
+import { L as LanguageModel } from '../../types-D774b0dg.mjs';
+import { a as AnthropicProviderConfig, A as AIProvider } from '../../types-BvkiJ1dd.mjs';
 import 'zod';
-import '../../base-tNgbBaSo.mjs';
+import '../../base-BYQKp9TW.mjs';
 
 /**
  * Anthropic Provider - Modern Pattern

package/dist/providers/anthropic/index.d.ts

@@ -1,7 +1,7 @@
-import { L as LanguageModel } from '../../types-BkQCSiIt.js';
-import { a as AnthropicProviderConfig, A as AIProvider } from '../../types-BQ31QIsA.js';
+import { L as LanguageModel } from '../../types-D774b0dg.js';
+import { a as AnthropicProviderConfig, A as AIProvider } from '../../types-TMilS-Dz.js';
 import 'zod';
-import '../../base-C58Dsr9p.js';
+import '../../base-Cxq3ni0t.js';
 
 /**
  * Anthropic Provider - Modern Pattern

package/dist/providers/anthropic/index.js

@@ -89,6 +89,63 @@ function toAnthropicOutputConfig(rf) {
     }
   };
 }
+function toAnthropicMcp(mcpServers) {
+  if (!mcpServers || mcpServers.length === 0) {
+    return { mcpServers: [], tools: [], betas: [] };
+  }
+  const serverEntries = [];
+  const toolEntries = [];
+  for (const mcp of mcpServers) {
+    const authHeader = mcp.headers?.Authorization ?? mcp.headers?.authorization;
+    const token = authHeader?.replace(/^Bearer\s+/i, "");
+    serverEntries.push({
+      type: "url",
+      url: mcp.url,
+      name: mcp.label,
+      ...token ? { authorization_token: token } : {}
+    });
+    if (mcp.allowedTools && mcp.allowedTools.length > 0) {
+      toolEntries.push({
+        type: "mcp_toolset",
+        mcp_server_name: mcp.label,
+        configs: Object.fromEntries(
+          mcp.allowedTools.map((toolName) => [toolName, {}])
+        )
+      });
+    }
+  }
+  return {
+    mcpServers: serverEntries,
+    tools: toolEntries,
+    betas: ["mcp-client-2025-11-20"]
+  };
+}
+function isStringEffort(effort) {
+  return typeof effort === "string" && (effort === "minimal" || effort === "low" || effort === "medium" || effort === "high");
+}
+var ANTHROPIC_ADAPTIVE_MODELS = /(claude-opus-4-7|claude-opus-4-6|claude-sonnet-4-6)/i;
+function toAnthropicThinking(effort, modelId) {
+  if (!effort) return {};
+  if (typeof effort === "object" && "raw" in effort) {
+    return { thinking: effort.raw };
+  }
+  const isAdaptive = !!modelId && ANTHROPIC_ADAPTIVE_MODELS.test(modelId);
+  if (typeof effort === "object" && "budgetTokens" in effort) {
+    return {
+      thinking: { type: "enabled", budget_tokens: effort.budgetTokens }
+    };
+  }
+  if (!isStringEffort(effort)) return {};
+  if (isAdaptive) {
+    const mapped = effort === "minimal" ? "low" : effort;
+    return {
+      thinking: { type: "adaptive" },
+      outputConfigEffort: mapped
+    };
+  }
+  const budget = effort === "high" ? 16e3 : effort === "medium" ? 8e3 : effort === "low" ? 4e3 : 2048;
+  return { thinking: { type: "enabled", budget_tokens: budget } };
+}
 function hasMediaAttachments(message) {
   const attachments = message.metadata?.attachments;
   return attachments?.some(
@@ -924,36 +981,58 @@ var AnthropicAdapter = class {
     if (serverToolConfiguration) {
       options.server_tool_configuration = serverToolConfiguration;
     }
+    const modelForThinking = request.config?.model || this.model;
+    const thinkingTranslation = toAnthropicThinking(
+      request.config?.reasoningEffort,
+      modelForThinking
+    );
     const outputConfig = toAnthropicOutputConfig(responseFormat);
-    if (outputConfig) {
-      options.output_config = outputConfig;
+    if (outputConfig || thinkingTranslation.outputConfigEffort) {
+      options.output_config = {
+        ...outputConfig ?? {},
+        ...thinkingTranslation.outputConfigEffort ? { effort: thinkingTranslation.outputConfigEffort } : {}
+      };
     }
-    if (this.config.thinking?.type === "enabled") {
+    if (thinkingTranslation.thinking) {
+      options.thinking = thinkingTranslation.thinking;
+    } else if (this.config.thinking?.type === "enabled") {
       options.thinking = {
         type: "enabled",
         budget_tokens: this.config.thinking.budgetTokens || 1e4
       };
     }
-    return { options, messages };
+    const mcp = toAnthropicMcp(request.config?.mcpServers);
+    const betas = [];
+    if (mcp.mcpServers.length > 0) {
+      options.mcp_servers = mcp.mcpServers;
+      betas.push(...mcp.betas);
+      if (mcp.tools.length > 0) {
+        const existingTools = Array.isArray(options.tools) ? options.tools : [];
+        options.tools = [...existingTools, ...mcp.tools];
+      }
+    }
+    return { options, messages, betas };
   }
   /**
    * Non-streaming completion (for debugging/comparison with original studio-ai)
    */
   async complete(request) {
     const client = await this.getClient();
-    const { options } = this.buildRequestOptions(request);
+    const { options, betas } = this.buildRequestOptions(request);
     const nonStreamingOptions = {
       ...options,
       stream: false
     };
     try {
+      const finalOptions = betas.length > 0 ? { ...nonStreamingOptions, betas } : nonStreamingOptions;
+      const messagesApi = betas.length > 0 ? client.beta.messages : client.messages;
       logProviderPayload(
         "anthropic",
         "request payload",
-        nonStreamingOptions,
+        finalOptions,
         request.debug
       );
-      const response = await client.messages.create(nonStreamingOptions);
+      const response = await messagesApi.create(finalOptions);
       logProviderPayload(
         "anthropic",
         "response payload",
@@ -988,17 +1067,19 @@ var AnthropicAdapter = class {
   }
   async *stream(request) {
     const client = await this.getClient();
-    const { options } = this.buildRequestOptions(request);
+    const { options, betas } = this.buildRequestOptions(request);
     const messageId = generateMessageId();
     yield { type: "message:start", id: messageId };
     try {
+      const finalOptions = betas.length > 0 ? { ...options, betas } : options;
+      const streamApi = betas.length > 0 ? client.beta.messages : client.messages;
       logProviderPayload(
         "anthropic",
         "request payload",
-        options,
+        finalOptions,
         request.debug
       );
-      const stream = await client.messages.stream(options);
+      const stream = await streamApi.stream(finalOptions);
       let currentToolUse = null;
       let isInThinkingBlock = false;
       const collectedCitations = [];