@massa-ai/mcp-client 1.67.0 → 1.68.0

package/dist/config-cli.js

@@ -484,13 +484,14 @@ var init_inference_providers = __esm(() => {
       knownDimensions: KNOWN_EMBEDDING_DIMENSIONS,
       defaultModels: {
         embedding: "qwen3-embedding:0.6b",
-        instruct: "qwen3-vl:8b",
-        coding: "qwen2.5-coder:7b"
+        instruct: "qwen3.5:9b",
+        coding: "qwen3.5:9b"
       },
       appliesContextPerRequest: true,
       embedBatchSize: 64,
       supportsOllamaVersionProbe: true,
       injectsDisableThink: true,
+      injectsChatTemplateKwargs: false,
       requiresChatCompletionsApi: false,
       parseModelList: parseOllamaModelList
     },
@@ -510,8 +511,8 @@ var init_inference_providers = __esm(() => {
       },
       defaultModels: {
         embedding: "text-embedding-qwen3-embedding-0.6b",
-        instruct: "qwen3-vl-8b-instruct",
-        coding: "qwen2.5-coder-7b-instruct"
+        instruct: "qwen3.8-9b",
+        coding: "qwen3.8-9b"
       },
       mlxModels: {
         embedding: {
@@ -519,23 +520,24 @@ var init_inference_providers = __esm(() => {
           model: "qwen3-embedding-0.6b-dwq"
         },
         instruct: {
-          repo: "mlx-community/Qwen3-VL-8B-Instruct-4bit",
-          model: "qwen3-vl-8b-instruct"
+          repo: "keXjos/Qwen3.8-9B-mlx-4Bit",
+          model: "qwen3.8-9b"
         },
         coding: {
-          repo: "mlx-community/Qwen2.5-Coder-7B-Instruct-4bit",
-          model: "qwen2.5-coder-7b-instruct"
+          repo: "keXjos/Qwen3.8-9B-mlx-4Bit",
+          model: "qwen3.8-9b"
         }
       },
       ggufRepos: {
         embedding: "Qwen/Qwen3-Embedding-0.6B-GGUF",
-        instruct: "lmstudio-community/Qwen3-VL-8B-Instruct-GGUF",
-        coding: "lmstudio-community/Qwen2.5-Coder-7B-Instruct-GGUF"
+        instruct: "empero-ai/Qwen3.8-9B-Distill-GGUF",
+        coding: "empero-ai/Qwen3.8-9B-Distill-GGUF"
       },
       appliesContextPerRequest: false,
       embedBatchSize: 64,
       supportsOllamaVersionProbe: false,
       injectsDisableThink: false,
+      injectsChatTemplateKwargs: true,
       requiresChatCompletionsApi: true,
       parseModelList: parseLmStudioModelList
     }
@@ -52852,9 +52854,13 @@ function _setLlmEnabledForTesting(flag) {
 function _resolveLlmConfig(cfg, role, baseUrlOverride) {
   const baseUrl = baseUrlOverride ?? cfg?.baseUrl ?? INFERENCE_PROVIDERS.ollama.defaultLlmBaseUrl;
   const spec = resolveInferenceSpec(baseUrl);
-  const model = role === "code" ? cfg?.codeModel ?? spec.defaultModels.coding : cfg?.model ?? spec.defaultModels.instruct;
+  const instructModel = cfg?.model ?? spec.defaultModels.instruct;
+  const codeModel = cfg?.codeModel ?? spec.defaultModels.coding;
+  const model = role === "code" ? codeModel : instructModel;
   const temperature = role === "code" ? cfg?.codeTemperature ?? INFERENCE_ROLE_DEFAULTS.coding.temperature : cfg?.temperature ?? INFERENCE_ROLE_DEFAULTS.instruct.temperature;
-  const contextWindow = role === "code" ? cfg?.codeContextWindow ?? INFERENCE_ROLE_DEFAULTS.coding.contextWindow : cfg?.contextWindow ?? INFERENCE_ROLE_DEFAULTS.instruct.contextWindow;
+  const instructContext = cfg?.contextWindow ?? INFERENCE_ROLE_DEFAULTS.instruct.contextWindow;
+  const codeContext = cfg?.codeContextWindow ?? INFERENCE_ROLE_DEFAULTS.coding.contextWindow;
+  const contextWindow = instructModel === codeModel ? Math.max(instructContext, codeContext) : role === "code" ? codeContext : instructContext;
   return {
     baseUrl,
     apiKey: cfg?.apiKey ?? spec.id,
@@ -52903,13 +52909,13 @@ function resolveInferenceSpec(baseUrl) {
 function resolveProviderIdForLogging(baseUrl) {
   return resolveMatchedProviderSpec(baseUrl)?.id ?? "unknown";
 }
-function _wrapFetchDisableThink(baseFetch) {
+function _wrapFetchDisableThink(baseFetch, field = "think", value = false) {
   const wrapped = async (input, init) => {
     try {
       if (init?.body && typeof init.body === "string") {
         const parsed = JSON.parse(init.body);
-        if (parsed && typeof parsed === "object" && !("think" in parsed)) {
-          parsed.think = false;
+        if (parsed && typeof parsed === "object" && !(field in parsed)) {
+          parsed[field] = value;
           init = { ...init, body: JSON.stringify(parsed) };
         }
       }
@@ -52942,6 +52948,11 @@ function buildProvider(llm) {
   if (llm.disableThink && spec.injectsDisableThink) {
     fetchImpl = _wrapFetchDisableThink(fetchImpl ?? globalThis.fetch);
   }
+  if (llm.disableThink && spec.injectsChatTemplateKwargs) {
+    fetchImpl = _wrapFetchDisableThink(fetchImpl ?? globalThis.fetch, "chat_template_kwargs", {
+      enable_thinking: false
+    });
+  }
   const openai2 = createOpenAI({
     baseURL: llm.baseUrl,
     apiKey: llm.apiKey,

package/dist/index.js

@@ -25481,13 +25481,14 @@ var init_inference_providers = __esm(() => {
       knownDimensions: KNOWN_EMBEDDING_DIMENSIONS,
       defaultModels: {
         embedding: "qwen3-embedding:0.6b",
-        instruct: "qwen3-vl:8b",
-        coding: "qwen2.5-coder:7b"
+        instruct: "qwen3.5:9b",
+        coding: "qwen3.5:9b"
       },
       appliesContextPerRequest: true,
       embedBatchSize: 64,
       supportsOllamaVersionProbe: true,
       injectsDisableThink: true,
+      injectsChatTemplateKwargs: false,
       requiresChatCompletionsApi: false,
       parseModelList: parseOllamaModelList
     },
@@ -25507,8 +25508,8 @@ var init_inference_providers = __esm(() => {
       },
       defaultModels: {
         embedding: "text-embedding-qwen3-embedding-0.6b",
-        instruct: "qwen3-vl-8b-instruct",
-        coding: "qwen2.5-coder-7b-instruct"
+        instruct: "qwen3.8-9b",
+        coding: "qwen3.8-9b"
       },
       mlxModels: {
         embedding: {
@@ -25516,23 +25517,24 @@ var init_inference_providers = __esm(() => {
           model: "qwen3-embedding-0.6b-dwq"
         },
         instruct: {
-          repo: "mlx-community/Qwen3-VL-8B-Instruct-4bit",
-          model: "qwen3-vl-8b-instruct"
+          repo: "keXjos/Qwen3.8-9B-mlx-4Bit",
+          model: "qwen3.8-9b"
         },
         coding: {
-          repo: "mlx-community/Qwen2.5-Coder-7B-Instruct-4bit",
-          model: "qwen2.5-coder-7b-instruct"
+          repo: "keXjos/Qwen3.8-9B-mlx-4Bit",
+          model: "qwen3.8-9b"
         }
       },
       ggufRepos: {
         embedding: "Qwen/Qwen3-Embedding-0.6B-GGUF",
-        instruct: "lmstudio-community/Qwen3-VL-8B-Instruct-GGUF",
-        coding: "lmstudio-community/Qwen2.5-Coder-7B-Instruct-GGUF"
+        instruct: "empero-ai/Qwen3.8-9B-Distill-GGUF",
+        coding: "empero-ai/Qwen3.8-9B-Distill-GGUF"
       },
       appliesContextPerRequest: false,
       embedBatchSize: 64,
       supportsOllamaVersionProbe: false,
       injectsDisableThink: false,
+      injectsChatTemplateKwargs: true,
       requiresChatCompletionsApi: true,
       parseModelList: parseLmStudioModelList
     }
@@ -59411,9 +59413,13 @@ function _setLlmEnabledForTesting(flag) {
 function _resolveLlmConfig(cfg, role, baseUrlOverride) {
   const baseUrl = baseUrlOverride ?? cfg?.baseUrl ?? INFERENCE_PROVIDERS.ollama.defaultLlmBaseUrl;
   const spec = resolveInferenceSpec(baseUrl);
-  const model = role === "code" ? cfg?.codeModel ?? spec.defaultModels.coding : cfg?.model ?? spec.defaultModels.instruct;
+  const instructModel = cfg?.model ?? spec.defaultModels.instruct;
+  const codeModel = cfg?.codeModel ?? spec.defaultModels.coding;
+  const model = role === "code" ? codeModel : instructModel;
   const temperature = role === "code" ? cfg?.codeTemperature ?? INFERENCE_ROLE_DEFAULTS.coding.temperature : cfg?.temperature ?? INFERENCE_ROLE_DEFAULTS.instruct.temperature;
-  const contextWindow = role === "code" ? cfg?.codeContextWindow ?? INFERENCE_ROLE_DEFAULTS.coding.contextWindow : cfg?.contextWindow ?? INFERENCE_ROLE_DEFAULTS.instruct.contextWindow;
+  const instructContext = cfg?.contextWindow ?? INFERENCE_ROLE_DEFAULTS.instruct.contextWindow;
+  const codeContext = cfg?.codeContextWindow ?? INFERENCE_ROLE_DEFAULTS.coding.contextWindow;
+  const contextWindow = instructModel === codeModel ? Math.max(instructContext, codeContext) : role === "code" ? codeContext : instructContext;
   return {
     baseUrl,
     apiKey: cfg?.apiKey ?? spec.id,
@@ -59462,13 +59468,13 @@ function resolveInferenceSpec(baseUrl) {
 function resolveProviderIdForLogging(baseUrl) {
   return resolveMatchedProviderSpec(baseUrl)?.id ?? "unknown";
 }
-function _wrapFetchDisableThink(baseFetch) {
+function _wrapFetchDisableThink(baseFetch, field = "think", value = false) {
   const wrapped = async (input, init) => {
     try {
       if (init?.body && typeof init.body === "string") {
         const parsed = JSON.parse(init.body);
-        if (parsed && typeof parsed === "object" && !("think" in parsed)) {
-          parsed.think = false;
+        if (parsed && typeof parsed === "object" && !(field in parsed)) {
+          parsed[field] = value;
           init = { ...init, body: JSON.stringify(parsed) };
         }
       }
@@ -59501,6 +59507,11 @@ function buildProvider(llm) {
   if (llm.disableThink && spec.injectsDisableThink) {
     fetchImpl = _wrapFetchDisableThink(fetchImpl ?? globalThis.fetch);
   }
+  if (llm.disableThink && spec.injectsChatTemplateKwargs) {
+    fetchImpl = _wrapFetchDisableThink(fetchImpl ?? globalThis.fetch, "chat_template_kwargs", {
+      enable_thinking: false
+    });
+  }
   const openai2 = createOpenAI({
     baseURL: llm.baseUrl,
     apiKey: llm.apiKey,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@massa-ai/mcp-client",
-  "version": "1.67.0",
+  "version": "1.68.0",
   "description": "massa-ai MCP server - Semantic code search, memory, and context compression via Model Context Protocol",
   "author": "luizgmassa",
   "type": "module",
@@ -20,8 +20,8 @@
     "type-check": "tsc --noEmit"
   },
   "dependencies": {
-    "@massa-ai/core": "^1.67.0",
-    "@massa-ai/shared": "^1.67.0",
+    "@massa-ai/core": "^1.68.0",
+    "@massa-ai/shared": "^1.68.0",
     "@modelcontextprotocol/sdk": "^1.0.0"
   },
   "devDependencies": {