converse-mcp-server 2.26.1 → 2.27.2

package/docs/PROVIDERS.md

@@ -37,12 +37,13 @@ This guide documents all supported AI providers in the Converse MCP Server and t
 - **Get Key**: [console.anthropic.com](https://console.anthropic.com/)
 - **Environment Variable**: `ANTHROPIC_API_KEY`
 - **Supported Models**:
-  - `claude-opus-4-5-20250220` - Most capable with extended thinking (32K output)
-  - `claude-opus-4-1-20250805` - Highest intelligence with extended thinking (32K output)
-  - `claude-sonnet-4-5-20250929` - Enhanced reasoning with extended thinking (64K output)
-  - `claude-sonnet-4-20250514` - Balanced performance with extended thinking (64K output)
-  - `claude-haiku-4-5-20251001` - Fast and intelligent with extended thinking (64K output)
-  - `claude-3-opus-20240229`, `claude-3-sonnet-20240229`, `claude-3-haiku-20240307` - Previous generation
+  - `claude-fable-5` (alias `fable`) - Most capable model for demanding reasoning and long-horizon agentic work (1M context, 128K output)
+  - `claude-opus-4-8` (alias `opus`) - Most capable Opus for complex reasoning and agentic coding (128K output)
+  - `claude-opus-4-7`, `claude-opus-4-6` - Previous Opus generations with adaptive thinking (128K output)
+  - `claude-opus-4-5-20251101`, `claude-opus-4-1-20250805` - Legacy Opus models (64K / 32K output)
+  - `claude-sonnet-4-6` (alias `sonnet`) - Best combination of speed and intelligence with adaptive thinking (64K output)
+  - `claude-sonnet-4-5-20250929` - Legacy Sonnet (64K output)
+  - `claude-haiku-4-5-20251001` (alias `haiku`) - Fast and intelligent with extended thinking (64K output)
 
 ### Mistral
 - **API Key Format**: 32+ character string
@@ -160,7 +161,7 @@ gemini
   "name": "consensus",
   "arguments": {
     "prompt": "Should we use TypeScript for this component?",
-    "models": ["gemini", "gpt-5", "claude-sonnet-4"]
+    "models": ["gemini", "gpt-5", "claude-sonnet-4-6"]
   }
 }
 ```
@@ -177,6 +178,27 @@ gemini
 - **Model Routing**: `gemini` → CLI provider, specific names (e.g., `gemini-2.5-pro`) → API provider
 - **Models**: Only gemini-3-pro-preview vs full Gemini model family
 
+### Claude Agent SDK
+- **Authentication**: Claude Code CLI login (no API key needed)
+- **Setup Required**: Authenticate once with `claude login` (Claude Code CLI)
+- **Environment Variables**: None (uses Claude Code credentials)
+- **Supported Models**:
+  - `claude` (aliases: `claude-sdk`, `claude-code`) - Defaults to Claude Fable 5
+  - `claude:fable` - Claude Fable 5 explicitly
+  - `claude:opus` - Claude Opus 4.8
+  - Other `claude:`-prefixed names pass through to the SDK (e.g. `claude:claude-sonnet-4-6`)
+
+**Key Features:**
+- **Subscription Access**: Uses your Claude subscription instead of API credits
+- **Local File Access**: Reads files directly from the working directory
+- **Image Support**: Via the SDK's streaming input mode
+- **SDK-Managed Parameters**: `temperature`, `use_websearch`, and `reasoning_effort` are managed internally (ignored if specified)
+
+**Differences from Anthropic API Provider:**
+- **Authentication**: Claude Code login vs `ANTHROPIC_API_KEY`
+- **Billing**: Claude subscription vs pay-per-use API
+- **Model Routing**: `claude` and `claude:*` → SDK provider; specific names (e.g., `claude-fable-5`, `opus`, `sonnet`) → API provider
+
 ## Configuration Examples
 
 ### Basic Configuration (.env file)
@@ -243,7 +265,7 @@ All providers support streaming responses for real-time output.
 - **Google**:
   - Gemini 3.0 Pro: Thinking levels (low/high) via `reasoning_effort` - always enabled
   - Gemini 2.5 Pro/Flash: Thinking budget (token-based) via `reasoning_effort`
-- **Anthropic**: All Claude 4 series models support extended thinking with `reasoning_effort`
+- **Anthropic**: Claude Fable 5, Opus 4.6+, and Sonnet 4.6 use adaptive thinking (depth controlled by `reasoning_effort` via Anthropic's `effort` parameter); older Claude 4 models use budget-based extended thinking
 - **Codex**: Thread-based agentic reasoning with persistent context
 - **Others**: Standard inference only
 
@@ -257,29 +279,38 @@ When using the chat or consensus tools, specify models using their identifiers:
 
 ### Model Routing Logic
 
-1. **Simple Names**: Models without "/" are routed by keyword matching:
+1. **SDK Providers** (exact matches and prefixes, checked first):
+   - `codex` → Codex
+   - `gemini`, `gemini-cli` → Gemini CLI
+   - `claude`, `claude-sdk`, `claude-code` and any `claude:`-prefixed name (e.g., `claude:fable`, `claude:opus`) → Claude Agent SDK
+   - `copilot`, `copilot-sdk`, `github-copilot` and any `copilot:`-prefixed name (e.g., `copilot:codex`) → Copilot SDK
+
+2. **Simple Names**: Other models without "/" are routed by keyword matching:
    - Contains "gpt", "o1", "o3", "o4" → OpenAI
-   - Contains "claude", "opus", "sonnet", "haiku" → Anthropic
+   - Contains "claude", "fable", "opus", "sonnet", "haiku" → Anthropic
    - Contains "gemini", "flash", "pro" → Google
    - Contains "grok" → X.AI
    - Contains "mistral", "magistral" → Mistral
    - Contains "deepseek", "reasoner", "r1" → DeepSeek
    - Contains "qwen", "kimi", "k2" → OpenRouter
 
-2. **Slash Format**: Models with "/" check native providers first:
+3. **Slash Format**: Models with "/" check native providers first:
    - If exact model exists in a native provider → Routes to that provider
    - If not found in any native provider → Routes to OpenRouter
    - This allows using models like "anthropic/claude-3.5-sonnet" via OpenRouter
 
-3. **OpenRouter Auto**: Special aliases route to OpenRouter's auto-selection:
+4. **OpenRouter Auto**: Special aliases route to OpenRouter's auto-selection:
    - "openrouter/auto", "openrouter auto", "auto router", "auto-router"
 
 ```javascript
 // Chat tool examples
 {
   "model": "gpt-5",               // OpenAI (keyword match)
-  "model": "claude-opus-4",       // Anthropic (keyword match, auto-resolves)
-  "model": "sonnet",              // Anthropic (keyword match)
+  "model": "fable",               // Anthropic (keyword match -> claude-fable-5)
+  "model": "opus",                // Anthropic (keyword match -> claude-opus-4-8)
+  "model": "sonnet",              // Anthropic (keyword match -> claude-sonnet-4-6)
+  "model": "claude",              // Claude Agent SDK (defaults to Claude Fable 5)
+  "model": "claude:opus",         // Claude Agent SDK (Claude Opus 4.8)
   "model": "gemini-2.5-pro",      // Google (keyword match)
   "model": "grok-4",              // X.AI (keyword match)
   "model": "mistral-large",       // Mistral (keyword match)
@@ -293,7 +324,7 @@ When using the chat or consensus tools, specify models using their identifiers:
 {
   "models": [
     {"model": "o3"},
-    {"model": "claude-sonnet-4"},
+    {"model": "claude-sonnet-4-6"},
     {"model": "gemini-2.5-pro"}
   ]
 }
@@ -308,7 +339,8 @@ When using the chat or consensus tools, specify models using their identifiers:
 
 ### Model Not Found
 - Use exact model identifiers as listed above
-- Some providers support aliases (e.g., "claude" → "claude-sonnet-4-5")
+- Some providers support aliases (e.g., "fable" → "claude-fable-5", "opus" → "claude-opus-4-8")
+- Note: bare "claude" routes to the Claude Agent SDK provider, not the Anthropic API
 - Check provider documentation for model availability in your region
 
 ### Rate Limits

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "converse-mcp-server",
-  "version": "2.26.1",
+  "version": "2.27.2",
   "description": "Converse MCP Server - Converse with other LLMs with chat and consensus tools",
   "type": "module",
   "main": "src/index.js",
@@ -94,30 +94,30 @@
     ".env.example"
   ],
   "dependencies": {
-    "@anthropic-ai/claude-agent-sdk": "^0.3.152",
-    "@anthropic-ai/sdk": "^0.99.0",
-    "@github/copilot-sdk": "^0.3.0",
-    "@google/genai": "^2.7.0",
-    "@mistralai/mistralai": "^2.2.1",
+    "@anthropic-ai/claude-agent-sdk": "^0.3.169",
+    "@anthropic-ai/sdk": "^0.102.0",
+    "@github/copilot-sdk": "^1.0.0",
+    "@google/genai": "^2.8.0",
+    "@mistralai/mistralai": "^2.2.5",
     "@modelcontextprotocol/sdk": "^1.29.0",
-    "@openai/codex-sdk": "^0.133.0",
-    "ai": "^6.0.191",
+    "@openai/codex-sdk": "^0.138.0",
+    "ai": "^6.0.198",
     "ai-sdk-provider-gemini-cli": "^2.0.1",
     "cors": "^2.8.6",
     "dotenv": "^17.4.2",
     "express": "^5.2.1",
-    "lru-cache": "^11.5.0",
+    "lru-cache": "^11.5.1",
     "nanoid": "^5.1.11",
-    "openai": "^6.39.0",
+    "openai": "^6.42.0",
     "p-limit": "^7.3.0",
-    "vite": "^8.0.14"
+    "vite": "^8.0.16"
   },
   "devDependencies": {
-    "@vitest/coverage-v8": "^4.1.7",
+    "@vitest/coverage-v8": "^4.1.8",
     "cross-env": "^10.1.0",
-    "eslint": "^10.4.0",
+    "eslint": "^10.4.1",
     "prettier": "^3.8.3",
     "rimraf": "^6.1.3",
-    "vitest": "^4.1.7"
+    "vitest": "^4.1.8"
   }
 }

package/src/providers/anthropic.js

@@ -13,6 +13,32 @@ import { ProviderError, ErrorCodes, StopReasons } from './interface.js';
 
 // Define supported Claude models with their capabilities
 const SUPPORTED_MODELS = {
+  'claude-fable-5': {
+    modelName: 'claude-fable-5',
+    friendlyName: 'Claude Fable 5',
+    contextWindow: 1000000, // 1M context by default - no beta header required
+    maxOutputTokens: 128000,
+    supportsStreaming: true,
+    supportsImages: true,
+    supportsTemperature: false, // Fable 5 rejects temperature/top_p/top_k (400)
+    supportsWebSearch: false,
+    supportsThinking: true,
+    supportsAdaptiveThinking: true, // Adaptive thinking is the only thinking mode
+    timeout: 600000,
+    supportsEffort: true,
+    effortGA: true,
+    supportsCompaction: true,
+    description:
+      'Claude Fable 5 - Most capable model for the most demanding reasoning and long-horizon agentic work',
+    aliases: [
+      'claude-fable-5',
+      'claude-fable',
+      'claude-5-fable',
+      'fable-5',
+      'fable5',
+      'fable',
+    ],
+  },
   'claude-opus-4-8': {
     modelName: 'claude-opus-4-8',
     friendlyName: 'Claude Opus 4.8',
@@ -678,10 +704,11 @@ export const anthropicProvider = {
       }
     }
 
-    // Add temperature if specified
+    // Add temperature if specified and the model accepts it
+    // (Fable 5 removed sampling parameters entirely - sending them returns 400)
     // When legacy thinking (type: "enabled") is active, temperature must be 1
     // Adaptive thinking (type: "adaptive") does not have this constraint
-    if (temperature !== undefined) {
+    if (temperature !== undefined && modelConfig.supportsTemperature !== false) {
       if (
         requestPayload.thinking &&
         requestPayload.thinking.type === 'enabled'

package/src/providers/claude.js

@@ -16,20 +16,46 @@
 import { debugLog, debugError } from '../utils/console.js';
 import { ProviderError, ErrorCodes, StopReasons } from './interface.js';
 
+// Default underlying model when the request is just "claude" (or "claude:fable")
+const DEFAULT_SDK_MODEL = 'claude-fable-5';
+
 // Supported Claude SDK models with their configurations
 const SUPPORTED_MODELS = {
-  claude: {
-    modelName: 'claude',
-    friendlyName: 'Claude (via Agent SDK)',
+  fable: {
+    modelName: 'claude-fable-5',
+    friendlyName: 'Claude Fable 5 (via Agent SDK)',
+    contextWindow: 1000000,
+    maxOutputTokens: 128000,
+    supportsStreaming: true,
+    supportsImages: true, // Supported via streaming input mode
+    supportsTemperature: false, // SDK manages temperature internally
+    supportsWebSearch: false, // SDK accesses files directly, not web
+    timeout: 600000, // 10 minutes
+    description:
+      'Claude Fable 5 via Agent SDK (default) - requires claude login authentication',
+    aliases: [
+      'claude',
+      'claude-sdk',
+      'claude-code',
+      'claude:fable',
+      'claude-fable-5',
+      'claude-fable',
+      'fable-5',
+    ],
+  },
+  opus: {
+    modelName: 'claude-opus-4-8',
+    friendlyName: 'Claude Opus 4.8 (via Agent SDK)',
     contextWindow: 200000,
-    maxOutputTokens: 8192,
+    maxOutputTokens: 128000,
     supportsStreaming: true,
     supportsImages: true, // Supported via streaming input mode
     supportsTemperature: false, // SDK manages temperature internally
     supportsWebSearch: false, // SDK accesses files directly, not web
     timeout: 600000, // 10 minutes
-    description: 'Claude via Agent SDK - requires claude login authentication',
-    aliases: ['claude-sdk', 'claude-code'],
+    description:
+      'Claude Opus 4.8 via Agent SDK - requires claude login authentication',
+    aliases: ['claude:opus', 'claude-opus-4-8'],
   },
 };
 
@@ -81,6 +107,61 @@ async function getClaudeSDK() {
   }
 }
 
+/**
+ * Look up model config from SUPPORTED_MODELS by name or alias.
+ * Strips the claude: prefix first (e.g. "claude:opus" -> "opus").
+ */
+function findModelConfig(modelName) {
+  if (typeof modelName !== 'string') return null;
+
+  let name = modelName.trim();
+  if (name.toLowerCase().startsWith('claude:')) {
+    name = name.slice('claude:'.length).trim();
+  }
+  if (!name) return SUPPORTED_MODELS.fable;
+
+  const nameLower = name.toLowerCase();
+
+  if (SUPPORTED_MODELS[nameLower]) {
+    return SUPPORTED_MODELS[nameLower];
+  }
+
+  for (const config of Object.values(SUPPORTED_MODELS)) {
+    if (
+      config.aliases &&
+      config.aliases.some((alias) => alias.toLowerCase() === nameLower)
+    ) {
+      return config;
+    }
+  }
+
+  return null;
+}
+
+/**
+ * Resolve the requested model to the underlying SDK model ID.
+ * - "claude" (and bare "claude:") defaults to Claude Fable 5
+ * - "claude:fable" / "claude:opus" select the specific model
+ * - Unknown names are passed through (after prefix stripping) so users can
+ *   target any model ID the Agent SDK accepts (e.g. "claude:claude-sonnet-4-6")
+ */
+function resolveSdkModel(modelName) {
+  if (typeof modelName !== 'string' || !modelName.trim()) {
+    return DEFAULT_SDK_MODEL;
+  }
+
+  const config = findModelConfig(modelName);
+  if (config) {
+    return config.modelName;
+  }
+
+  let name = modelName.trim();
+  if (name.toLowerCase().startsWith('claude:')) {
+    name = name.slice('claude:'.length).trim();
+  }
+  return name || DEFAULT_SDK_MODEL;
+}
+
 /**
  * Convert message array to SDK input format
  * Claude SDK supports two modes:
@@ -209,7 +290,7 @@ async function* createSdkMessageGenerator(sdkMessage) {
  * @param {Function} queryFn - The SDK query function
  * @param {string|null} prompt - String prompt for single message mode, or null for streaming input mode
  * @param {Object|null} sdkMessage - SDK user message for streaming input mode (with images)
- * @param {Object} options - SDK options (cwd, etc.)
+ * @param {Object} options - SDK options (cwd, model, etc.)
  * @param {AbortSignal} signal - Abort signal for cancellation
  */
 async function* createStreamingGenerator(
@@ -223,7 +304,7 @@ async function* createStreamingGenerator(
     // Build query options
     // Use higher maxTurns to allow for file reading operations
     const queryOptions = {
-      model: 'claude-opus-4-8', // Use Opus 4.8 for best quality
+      model: options.model || DEFAULT_SDK_MODEL,
       maxTurns: 20, // Allow multiple turns for file operations
       permissionMode: 'bypassPermissions', // Don't prompt for permissions
     };
@@ -259,14 +340,14 @@ async function* createStreamingGenerator(
     });
 
     let _sessionId = null;
-    let _modelUsed = 'claude';
+    let _modelUsed = queryOptions.model;
     let _accumulatedContent = '';
 
     // Yield start event
     yield {
       type: 'start',
       provider: 'claude',
-      model: 'claude',
+      model: queryOptions.model,
     };
 
     // Iterate over SDK messages
@@ -281,7 +362,7 @@ async function* createStreamingGenerator(
       case 'system':
         if (message.subtype === 'init') {
           _sessionId = message.session_id;
-          _modelUsed = message.model || 'claude';
+          _modelUsed = message.model || queryOptions.model;
           debugLog(
             `[Claude SDK] Session initialized: ${_sessionId}, model: ${_modelUsed}`,
           );
@@ -414,9 +495,14 @@ export const claudeProvider = {
         debugLog('[Claude SDK] Using streaming input mode for image support');
       }
 
+      // Resolve requested model (claude/claude:fable -> claude-fable-5, claude:opus -> claude-opus-4-8)
+      const sdkModel = resolveSdkModel(model);
+      debugLog(`[Claude SDK] Resolved model "${model}" -> "${sdkModel}"`);
+
       // Build SDK options
       const sdkOptions = {
         cwd: config.server?.client_cwd || process.cwd(),
+        model: sdkModel,
       };
 
       // Streaming mode
@@ -459,7 +545,7 @@ export const claudeProvider = {
         rawResponse: { content, usage },
         metadata: {
           provider: 'claude',
-          model,
+          model: sdkModel,
           usage: usage
             ? {
               input_tokens: usage.input_tokens || 0,
@@ -546,25 +632,9 @@ export const claudeProvider = {
 
   /**
    * Get model configuration for specific model
+   * Handles claude: prefixed names (e.g. "claude:opus", "claude:fable")
    */
   getModelConfig(modelName) {
-    const modelNameLower = modelName.toLowerCase();
-
-    // Check exact match
-    if (SUPPORTED_MODELS[modelNameLower]) {
-      return SUPPORTED_MODELS[modelNameLower];
-    }
-
-    // Check aliases
-    for (const [_name, config] of Object.entries(SUPPORTED_MODELS)) {
-      if (
-        config.aliases &&
-        config.aliases.some((alias) => alias.toLowerCase() === modelNameLower)
-      ) {
-        return config;
-      }
-    }
-
-    return null;
+    return findModelConfig(modelName);
   },
 };

package/src/providers/copilot.js

@@ -7,12 +7,11 @@
  * Key differences from traditional providers:
  * - Uses GitHub Copilot CLI subscription authentication - NOT API keys
  * - Manages a singleton CopilotClient (spawns CLI process via JSON-RPC)
- * - Creates a fresh CopilotSession per request, destroyed after each request
+ * - Creates a fresh CopilotSession per request, disconnected after each request
  * - Bridges SDK push-based events to pull-based async generator for streaming
  * - Requires GitHub CLI authenticated (gh auth login) with active Copilot subscription
  */
 
-import { register } from 'node:module';
 import { debugLog, debugError } from '../utils/console.js';
 import { ProviderError, ErrorCodes, StopReasons } from './interface.js';
 
@@ -317,28 +316,10 @@ function isCopilotSDKAvailable() {
   return _sdkAvailable;
 }
 
-/**
- * Register a module resolution hook to fix the extensionless
- * "vscode-jsonrpc/node" import in @github/copilot-sdk >=0.1.29.
- * The SDK's session.js imports "vscode-jsonrpc/node" without a .js extension,
- * which fails under Node.js strict ESM resolution. The hook rewrites it to
- * "vscode-jsonrpc/node.js". Runs once, works regardless of package manager.
- */
-let _hookRegistered = false;
-function ensureJsonrpcResolveHook() {
-  if (_hookRegistered) return;
-  _hookRegistered = true;
-  register(`data:text/javascript,${encodeURIComponent(
-    'export function resolve(s,c,n){' +
-    'return s==="vscode-jsonrpc/node"?n("vscode-jsonrpc/node.js",c):n(s,c);}',
-  )}`);
-}
-
 /**
  * Dynamically import Copilot SDK (lazy loading)
  */
 async function getCopilotSDK() {
-  ensureJsonrpcResolveHook();
   try {
     const { CopilotClient } = await import('@github/copilot-sdk');
     return CopilotClient;
@@ -370,14 +351,13 @@ async function getCopilotClient(cwd) {
 
   clientInitPromise = (async () => {
     const CopilotClient = await getCopilotSDK();
+    const workingDirectory = cwd || process.cwd();
     clientInstance = new CopilotClient({
-      autoStart: true,
-      autoRestart: true,
       useLoggedInUser: true,
-      cwd: cwd || process.cwd(),
+      workingDirectory,
     });
     await clientInstance.start();
-    debugLog('[Copilot SDK] Client started (cwd: %s)', clientInstance.options?.cwd || cwd);
+    debugLog('[Copilot SDK] Client started (cwd: %s)', workingDirectory);
     return clientInstance;
   })();
 
@@ -808,9 +788,9 @@ async function* createStreamingGenerator(client, prompt, options, signal, config
     }
   } finally {
     try {
-      await session.destroy();
-    } catch (destroyError) {
-      debugError('[Copilot SDK] Session destroy error', destroyError);
+      await session.disconnect();
+    } catch (disconnectError) {
+      debugError('[Copilot SDK] Session disconnect error', disconnectError);
     }
   }
 }

package/src/tools/chat.js

@@ -599,6 +599,12 @@ export function mapModelToProvider(model, providers) {
     return 'claude';
   }
 
+  // Check claude: prefix (e.g., claude:fable, claude:opus) - routes to SDK provider
+  // Must be before keyword matching to prevent misrouting to Anthropic API
+  if (modelLower.startsWith('claude:')) {
+    return 'claude';
+  }
+
   // Check Copilot SDK (exact match only - routes to SDK provider)
   if (
     modelLower === 'copilot' ||
@@ -673,6 +679,7 @@ export function mapModelToProvider(model, providers) {
   // Anthropic models
   if (
     modelLower.includes('claude') ||
+    modelLower.includes('fable') ||
     modelLower.includes('opus') ||
     modelLower.includes('sonnet') ||
     modelLower.includes('haiku')
@@ -1149,7 +1156,7 @@ chatTool.inputSchema = {
     model: {
       type: 'string',
       description:
-        'AI model to use. Examples: "auto" (recommended), "codex", "gemini", "claude", "copilot", "copilot:codex". Defaults to auto-selection.',
+        'AI model to use. Examples: "auto" (recommended), "codex", "gemini", "claude", "claude:fable", "claude:opus", "copilot", "copilot:codex". Defaults to auto-selection.',
     },
     files: {
       type: 'array',

package/src/tools/consensus.js

@@ -803,6 +803,12 @@ function mapModelToProvider(model, providers) {
     return 'claude';
   }
 
+  // Check claude: prefix (e.g., claude:fable, claude:opus) - routes to SDK provider
+  // Must be before keyword matching to prevent misrouting to Anthropic API
+  if (modelLower.startsWith('claude:')) {
+    return 'claude';
+  }
+
   // Check Copilot SDK (exact match only - routes to SDK provider)
   if (
     modelLower === 'copilot' ||
@@ -877,6 +883,7 @@ function mapModelToProvider(model, providers) {
   // Anthropic models
   if (
     modelLower.includes('claude') ||
+    modelLower.includes('fable') ||
     modelLower.includes('opus') ||
     modelLower.includes('sonnet') ||
     modelLower.includes('haiku')
@@ -1705,7 +1712,7 @@ consensusTool.inputSchema = {
       items: { type: 'string' },
       minItems: 1,
       description:
-        'List of models to consult. Examples: ["codex", "gemini", "claude", "copilot", "copilot:codex"]',
+        'List of models to consult. Examples: ["codex", "gemini", "claude", "claude:opus", "copilot", "copilot:codex"]',
     },
     files: {
       type: 'array',

package/src/utils/modelRouting.js

@@ -89,6 +89,12 @@ export function mapModelToProvider(model, providers) {
     return 'claude';
   }
 
+  // Check claude: prefix (e.g., claude:fable, claude:opus) - routes to SDK provider
+  // Must be before keyword matching to prevent misrouting to Anthropic API
+  if (modelLower.startsWith('claude:')) {
+    return 'claude';
+  }
+
   // Check Copilot SDK (exact match only - routes to SDK provider)
   if (
     modelLower === 'copilot' ||
@@ -163,6 +169,7 @@ export function mapModelToProvider(model, providers) {
   // Anthropic models
   if (
     modelLower.includes('claude') ||
+    modelLower.includes('fable') ||
     modelLower.includes('opus') ||
     modelLower.includes('sonnet') ||
     modelLower.includes('haiku')