converse-mcp-server 2.28.0 → 2.29.0

package/.env.example

@@ -77,9 +77,9 @@ OPENROUTER_REFERER=https://github.com/FallDownTheSystem/converse
 # WARNING: Interactive policies may cause hangs in server/headless mode
 # CODEX_APPROVAL_POLICY=never
 
-# Default Codex model (default: gpt-5.5)
-# Options: gpt-5.5, gpt-5.3-codex, gpt-5-codex
-# CODEX_MODEL=gpt-5.5
+# Default Codex model (default: gpt-5.6)
+# Options: gpt-5.6, gpt-5.6-terra, gpt-5.6-luna, gpt-5.5, gpt-5.3-codex
+# CODEX_MODEL=gpt-5.6
 
 # ============================================
 # Server Configuration

package/docs/API.md

@@ -516,11 +516,13 @@ When complete, `check_status` for the continuation_id renders the full lap trans
 
 | Model | Context | Tokens | Features | Use Cases |
 |-------|---------|--------|----------|-----------|
-| `gpt-5.1` | 1M | 128K | Latest GPT | Multimodal, general purpose |
-| `gpt-5` | 1M | 64K | Advanced | Complex reasoning, analysis |
-| `gpt-5-mini` | 1M | 64K | Fast | Balanced performance/speed |
-| `gpt-5-nano` | 1M | 64K | Ultra-fast | Quick responses, simple queries |
-| `gpt-5-pro` | 1M | 128K | Pro tier | Extended capabilities |
+| `gpt-5.6-sol` (`gpt-5.6`, `gpt-5`, `sol`) | 1M | 128K | Flagship, default | Frontier reasoning, coding, agentic workflows |
+| `gpt-5.6-terra` (`terra`) | 400K | 128K | Lower cost | Strong performance at half the flagship price |
+| `gpt-5.6-luna` (`luna`) | 400K | 128K | Fastest | High-volume, latency-sensitive workloads |
+| `gpt-5.4` | 1M | 128K | Previous flagship | Complex reasoning, analysis |
+| `gpt-5.4-pro` | 1M | 272K | Pro tier | Extended capabilities (expensive) |
+| `gpt-5-mini` | 400K | 128K | Fast | Balanced performance/speed |
+| `gpt-5-nano` | 400K | 128K | Ultra-fast | Quick responses, simple queries |
 | `o3` | 200K | 100K | Reasoning | Logic, analysis, complex problems |
 | `o3-pro` | 200K | 100K | Extended reasoning | Deep analysis |
 | `o4-mini` | 200K | 100K | Fast reasoning | General purpose, rapid reasoning |

package/docs/PROVIDERS.md

@@ -9,7 +9,10 @@ This guide documents all supported AI providers in the Converse MCP Server and t
 - **Get Key**: [platform.openai.com/api-keys](https://platform.openai.com/api-keys)
 - **Environment Variable**: `OPENAI_API_KEY`
 - **Supported Models**:
-  - `gpt-5.1`, `gpt-5`, `gpt-5-mini`, `gpt-5-nano`, `gpt-5-pro` - GPT-5 family with advanced reasoning
+  - `gpt-5.6-sol` (aliases: `gpt-5.6`, `gpt-5`, `sol`) - Flagship GPT-5.6, the default OpenAI model
+  - `gpt-5.6-terra` (alias: `terra`) - Lower-cost GPT-5.6, competitive with GPT-5.5
+  - `gpt-5.6-luna` (alias: `luna`) - Fastest, most affordable GPT-5.6
+  - `gpt-5.4`, `gpt-5.4-mini`, `gpt-5.4-nano`, `gpt-5.4-pro`, `gpt-5-mini`, `gpt-5-nano` - GPT-5.4/GPT-5 family
   - `o3`, `o3-pro`, `o4-mini` - Advanced reasoning models
   - `gpt-4.1` - Large context (1M tokens)
   - `o3-deep-research`, `o4-mini-deep-research` - Deep research models
@@ -92,8 +95,9 @@ This guide documents all supported AI providers in the Converse MCP Server and t
   - `CODEX_SANDBOX_MODE` - Filesystem access control (default: read-only)
   - `CODEX_SKIP_GIT_CHECK` - Skip Git repository validation (default: true)
   - `CODEX_APPROVAL_POLICY` - Command approval behavior (default: never)
+  - `CODEX_MODEL` - Underlying model for Codex sessions (default: gpt-5.6; e.g. gpt-5.6-terra, gpt-5.6-luna, gpt-5.5)
 - **Supported Models**:
-  - `codex` - OpenAI Codex agentic coding assistant
+  - `codex` - OpenAI Codex agentic coding assistant (GPT-5.6 by default)
   - Thread-based sessions with persistent context
   - Direct filesystem access from working directory
   - Typical response time: 6-20 seconds (longer for complex tasks)

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "converse-mcp-server",
-  "version": "2.28.0",
+  "version": "2.29.0",
   "description": "Converse MCP Server - Converse with other LLMs with chat and consensus tools",
   "type": "module",
   "main": "src/index.js",
@@ -93,29 +93,29 @@
     ".env.example"
   ],
   "dependencies": {
-    "@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",
+    "@anthropic-ai/claude-agent-sdk": "^0.3.201",
+    "@anthropic-ai/sdk": "^0.110.0",
+    "@github/copilot-sdk": "^1.0.5",
+    "@google/genai": "^2.10.0",
     "@lydell/node-pty": "1.2.0-beta.12",
-    "@mistralai/mistralai": "^2.2.5",
+    "@mistralai/mistralai": "^2.4.1",
     "@modelcontextprotocol/sdk": "^1.29.0",
-    "@openai/codex-sdk": "^0.138.0",
-    "ai": "^6.0.198",
+    "@openai/codex-sdk": "^0.142.5",
+    "ai": "^6.0.208",
     "cors": "^2.8.6",
     "dotenv": "^17.4.2",
     "express": "^5.2.1",
     "lru-cache": "^11.5.1",
-    "nanoid": "^5.1.11",
-    "openai": "^6.42.0",
+    "nanoid": "^5.1.16",
+    "openai": "^6.45.0",
     "p-limit": "^7.3.0",
-    "vite": "^8.0.16"
+    "vite": "^8.1.3"
   },
   "devDependencies": {
-    "@vitest/coverage-v8": "^4.1.8",
+    "@vitest/coverage-v8": "^4.1.9",
     "cross-env": "^10.1.0",
-    "eslint": "^10.4.1",
+    "eslint": "^10.6.0",
     "rimraf": "^6.1.3",
-    "vitest": "^4.1.8"
+    "vitest": "^4.1.9"
   }
 }

package/src/config.js

@@ -279,9 +279,9 @@ const CONFIG_SCHEMA = {
     },
     CODEX_MODEL: {
       type: 'string',
-      default: 'gpt-5.5',
+      default: 'gpt-5.6',
       description:
-        'Default Codex model (e.g., gpt-5.5, gpt-5.3-codex, gpt-5-codex)',
+        'Default Codex model (e.g., gpt-5.6, gpt-5.6-terra, gpt-5.6-luna, gpt-5.5)',
     },
 
     // Copilot configuration
@@ -296,6 +296,12 @@ const CONFIG_SCHEMA = {
       description:
         'Default model for Copilot SDK sessions (e.g., gpt-5, claude-sonnet-4.5)',
     },
+    COPILOT_CLI_PATH: {
+      type: 'string',
+      required: false,
+      description:
+        'Explicit path to the Copilot CLI runtime (index.js or copilot binary). Overrides automatic resolution.',
+    },
   },
 
   // MCP configuration

package/src/providers/codex.js

@@ -21,7 +21,7 @@ import { normalizeExtendedPath } from '../utils/pathUtils.js';
 const SUPPORTED_MODELS = {
   codex: {
     modelName: 'codex',
-    friendlyName: 'OpenAI Codex (GPT-5.5)',
+    friendlyName: 'OpenAI Codex (GPT-5.6)',
     contextWindow: 400000,
     maxOutputTokens: 128000,
     supportsStreaming: true,
@@ -30,7 +30,7 @@ const SUPPORTED_MODELS = {
     supportsWebSearch: false, // Codex accesses files directly, not web
     timeout: 600000, // 10 minutes
     description:
-      'OpenAI Codex agentic coding assistant with local file access and tool execution (GPT-5.5)',
+      'OpenAI Codex agentic coding assistant with local file access and tool execution (GPT-5.6)',
     aliases: [
       'gpt-5-codex',
       'gpt5-codex',
@@ -39,6 +39,8 @@ const SUPPORTED_MODELS = {
       'gpt5.3-codex',
       'gpt-5.5',
       'gpt5.5',
+      'gpt-5.6-codex',
+      'gpt5.6-codex',
     ],
   },
 };

package/src/providers/copilot.js

@@ -12,6 +12,9 @@
  * - Requires GitHub CLI authenticated (gh auth login) with active Copilot subscription
  */
 
+import { existsSync } from 'node:fs';
+import { delimiter, dirname, join } from 'node:path';
+import { fileURLToPath } from 'node:url';
 import { debugLog, debugError } from '../utils/console.js';
 import { ProviderError, ErrorCodes, StopReasons } from './interface.js';
 
@@ -321,8 +324,10 @@ function isCopilotSDKAvailable() {
  */
 async function getCopilotSDK() {
   try {
-    const { CopilotClient } = await import('@github/copilot-sdk');
-    return CopilotClient;
+    const { CopilotClient, RuntimeConnection } = await import(
+      '@github/copilot-sdk'
+    );
+    return { CopilotClient, RuntimeConnection };
   } catch (error) {
     throw new CopilotProviderError(
       `Copilot SDK import failed: ${error.message}`,
@@ -332,6 +337,78 @@ async function getCopilotSDK() {
   }
 }
 
+/**
+ * Locate a runnable `copilot` executable on PATH (e.g. a winget or global
+ * install). On Windows only real executables (.exe/.com) qualify — .cmd/.bat
+ * shims can't be spawned without a shell, and the SDK spawns the CLI without
+ * one, so handing it a shim would fail.
+ */
+function findCopilotBinaryOnPath() {
+  const pathDirs = (process.env.PATH || process.env.Path || '')
+    .split(delimiter)
+    .filter(Boolean);
+  const exts = process.platform === 'win32' ? ['.exe', '.com'] : [''];
+  for (const dir of pathDirs) {
+    for (const ext of exts) {
+      const candidate = join(dir, `copilot${ext}`);
+      if (existsSync(candidate)) return candidate;
+    }
+  }
+  return null;
+}
+
+/**
+ * Resolve the Copilot CLI runtime path deterministically.
+ *
+ * The SDK's built-in resolver (getBundledCliPath) reconstructs
+ * `<@github/copilot root>/index.js` from the package's `./sdk` export. That
+ * assumption breaks across layouts — pnpm stores with multiple versions, global
+ * installs, and loader-style `@github/copilot` variants that ship no root
+ * `index.js` — and there is no escape hatch wired up, so a wrong/missing
+ * bundled copy takes down the whole provider (the "Copilot CLI not found at
+ * ...@github\index.js" failure). We own discovery here and hand the SDK an
+ * explicit, verified path so that entire class of resolution failures can't
+ * occur.
+ *
+ * Precedence: explicit override → bundled npm CLI (index.js, run via node) →
+ * `copilot` on PATH. Returns null to let the SDK fall back to its own
+ * resolution as a last resort (preserves behavior on flat installs).
+ */
+function resolveCopilotCliPath(config) {
+  // 1. Explicit override (COPILOT_CLI_PATH via config or env)
+  const override =
+    config?.providers?.copilotclipath || process.env.COPILOT_CLI_PATH;
+  if (override && existsSync(override)) {
+    debugLog('[Copilot SDK] Using configured CLI path: %s', override);
+    return override;
+  }
+
+  // 2. Bundled npm CLI: resolve the @github/copilot/sdk export, walk to the
+  // package root, and use its index.js. A .js path is spawned via `node`, which
+  // also sidesteps Windows .cmd/.bat shim issues.
+  try {
+    const sdkEntry = fileURLToPath(import.meta.resolve('@github/copilot/sdk'));
+    const candidate = join(dirname(dirname(sdkEntry)), 'index.js');
+    if (existsSync(candidate)) {
+      debugLog('[Copilot SDK] Using bundled CLI: %s', candidate);
+      return candidate;
+    }
+  } catch {
+    // @github/copilot not resolvable from here — fall through.
+  }
+
+  // 3. Standalone `copilot` binary on PATH.
+  const binary = findCopilotBinaryOnPath();
+  if (binary) {
+    debugLog('[Copilot SDK] Using copilot binary from PATH: %s', binary);
+    return binary;
+  }
+
+  // 4. Nothing found — defer to the SDK's own resolution.
+  debugLog('[Copilot SDK] No CLI path resolved — deferring to SDK default');
+  return null;
+}
+
 // Module-level singleton client
 let clientInstance = null;
 let clientInitPromise = null;
@@ -340,7 +417,7 @@ let clientInitPromise = null;
  * Get or create the singleton CopilotClient
  * The client manages the CLI process lifecycle via JSON-RPC
  */
-async function getCopilotClient(cwd) {
+async function getCopilotClient(cwd, config) {
   if (clientInstance) {
     return clientInstance;
   }
@@ -350,12 +427,19 @@ async function getCopilotClient(cwd) {
   }
 
   clientInitPromise = (async () => {
-    const CopilotClient = await getCopilotSDK();
+    const { CopilotClient, RuntimeConnection } = await getCopilotSDK();
     const workingDirectory = cwd || process.cwd();
-    clientInstance = new CopilotClient({
+    const clientOptions = {
       useLoggedInUser: true,
       workingDirectory,
-    });
+    };
+
+    const cliPath = resolveCopilotCliPath(config);
+    if (cliPath) {
+      clientOptions.connection = RuntimeConnection.forStdio({ path: cliPath });
+    }
+
+    clientInstance = new CopilotClient(clientOptions);
     await clientInstance.start();
     debugLog('[Copilot SDK] Client started (cwd: %s)', workingDirectory);
     return clientInstance;
@@ -795,7 +879,7 @@ async function* createStreamingGenerator(client, prompt, options, signal, config
   }
 }
 
-export { resolveModelAlias, resolveSessionModel };
+export { resolveModelAlias, resolveSessionModel, resolveCopilotCliPath };
 
 /**
  * Copilot SDK Provider Implementation
@@ -831,7 +915,7 @@ export const copilotProvider = {
     }
     try {
       const cwd = config.server?.client_cwd || process.cwd();
-      const client = await getCopilotClient(cwd);
+      const client = await getCopilotClient(cwd, config);
       const prompt = convertMessagesToPrompt(messages);
 
       const sessionModel = resolveSessionModel(model, config);

package/src/providers/openai.js

@@ -10,6 +10,64 @@ import { debugLog, debugError } from '../utils/console.js';
 
 // Define supported models with their capabilities
 const SUPPORTED_MODELS = {
+  'gpt-5.6-sol': {
+    modelName: 'gpt-5.6-sol',
+    friendlyName: 'OpenAI (GPT-5.6 Sol)',
+    contextWindow: 1000000,
+    maxOutputTokens: 128000,
+    supportsStreaming: true,
+    supportsImages: true,
+    supportsTemperature: false,
+    supportsWebSearch: true,
+    supportsResponsesAPI: true,
+    supportsNoneReasoningEffort: true,
+    timeout: 3600000, // 1 hour
+    description:
+      'Flagship GPT-5.6 model (1M context, 128K output) - Frontier reasoning, coding, agentic workflows. Most token-efficient flagship',
+    aliases: [
+      'gpt-5.6',
+      'gpt5.6',
+      'gpt 5.6',
+      'gpt-5',
+      'gpt5',
+      'gpt 5',
+      'sol',
+      'gpt-5.6sol',
+      'gpt 5.6 sol',
+    ],
+  },
+  'gpt-5.6-terra': {
+    modelName: 'gpt-5.6-terra',
+    friendlyName: 'OpenAI (GPT-5.6 Terra)',
+    contextWindow: 400000,
+    maxOutputTokens: 128000,
+    supportsStreaming: true,
+    supportsImages: true,
+    supportsTemperature: false,
+    supportsWebSearch: true,
+    supportsResponsesAPI: true,
+    supportsNoneReasoningEffort: true,
+    timeout: 1800000, // 30 minutes
+    description:
+      'Lower-cost GPT-5.6 (400K context, 128K output) - Performance competitive with GPT-5.5 at half the flagship price',
+    aliases: ['gpt5.6-terra', 'gpt-5.6terra', 'gpt 5.6 terra', 'terra'],
+  },
+  'gpt-5.6-luna': {
+    modelName: 'gpt-5.6-luna',
+    friendlyName: 'OpenAI (GPT-5.6 Luna)',
+    contextWindow: 400000,
+    maxOutputTokens: 128000,
+    supportsStreaming: true,
+    supportsImages: true,
+    supportsTemperature: false,
+    supportsWebSearch: true,
+    supportsResponsesAPI: true,
+    supportsNoneReasoningEffort: true,
+    timeout: 600000, // 10 minutes
+    description:
+      'Fastest, most affordable GPT-5.6 (400K context, 128K output) - High-volume, latency-sensitive workloads',
+    aliases: ['gpt5.6-luna', 'gpt-5.6luna', 'gpt 5.6 luna', 'luna'],
+  },
   'gpt-5.4': {
     modelName: 'gpt-5.4',
     friendlyName: 'OpenAI (GPT-5.4)',
@@ -25,9 +83,6 @@ const SUPPORTED_MODELS = {
     description:
       'Latest flagship model (1M context, 128K output) - Superior reasoning, coding, agentic workflows, computer use. Most token-efficient reasoning model',
     aliases: [
-      'gpt-5',
-      'gpt5',
-      'gpt 5',
       'gpt5.4',
       'gpt 5.4',
     ],
@@ -270,6 +325,22 @@ function resolveModelName(modelName) {
   return modelName;
 }
 
+/**
+ * Resolve the reasoning effort actually sent to the API for a given model.
+ * GPT-5 Pro models only accept 'high'. GPT-5.6 models dropped 'minimal'
+ * (supported efforts: none, low, medium, high, xhigh, max), so 'minimal'
+ * maps to the closest supported value.
+ */
+function resolveReasoningEffort(resolvedModel, reasoningEffort) {
+  if (resolvedModel.endsWith('-pro') && resolvedModel.startsWith('gpt-5')) {
+    return 'high';
+  }
+  if (resolvedModel.startsWith('gpt-5.6') && reasoningEffort === 'minimal') {
+    return 'low';
+  }
+  return reasoningEffort;
+}
+
 /**
  * Validate OpenAI API key format
  */
@@ -393,7 +464,7 @@ export const openaiProvider = {
    */
   async invoke(messages, options = {}) {
     const {
-      model = 'gpt-4o-mini',
+      model = 'gpt-5.6',
       temperature = 0.7,
       maxTokens = null,
       stream = false,
@@ -470,11 +541,8 @@ export const openaiProvider = {
         (resolvedModel.startsWith('o3') || resolvedModel.startsWith('gpt-5')) &&
         reasoning_effort
       ) {
-        // GPT-5 Pro only supports 'high' reasoning effort
-        const effectiveEffort =
-          resolvedModel.endsWith('-pro') && resolvedModel.startsWith('gpt-5') ? 'high' : reasoning_effort;
         requestPayload.reasoning = {
-          effort: effectiveEffort,
+          effort: resolveReasoningEffort(resolvedModel, reasoning_effort),
           summary: 'auto', // Enable reasoning summaries
         };
       }
@@ -510,10 +578,10 @@ export const openaiProvider = {
         (resolvedModel.startsWith('o3') || resolvedModel.startsWith('gpt-5')) &&
         reasoning_effort
       ) {
-        // GPT-5 Pro only supports 'high' reasoning effort
-        const effectiveEffort =
-          resolvedModel.endsWith('-pro') && resolvedModel.startsWith('gpt-5') ? 'high' : reasoning_effort;
-        requestPayload.reasoning_effort = effectiveEffort;
+        requestPayload.reasoning_effort = resolveReasoningEffort(
+          resolvedModel,
+          reasoning_effort,
+        );
       }
 
       // Add verbosity for GPT-5 models

package/src/utils/modelRouting.js

@@ -18,7 +18,7 @@ export function getDefaultModelForProvider(providerName) {
     'gemini-cli': 'gemini',
     claude: 'claude',
     copilot: 'copilot',
-    openai: 'gpt-5',
+    openai: 'gpt-5.6',
     xai: 'grok-4-0709',
     google: 'gemini-pro',
     anthropic: 'claude-sonnet-4-20250514',
@@ -27,7 +27,7 @@ export function getDefaultModelForProvider(providerName) {
     openrouter: 'qwen/qwen3-coder',
   };
 
-  return defaults[providerName] || 'gpt-5';
+  return defaults[providerName] || 'gpt-5.6';
 }
 
 /**