converse-mcp-server 1.5.1 → 1.5.3

package/LICENSE

@@ -0,0 +1,19 @@
+Copyright 2025 Converse MCP Server Contributors
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

package/README.md

@@ -1,20 +1,22 @@
 # Converse MCP Server
 
+[![npm version](https://img.shields.io/npm/v/converse-mcp-server.svg)](https://www.npmjs.com/package/converse-mcp-server)
+
 A simplified, functional Node.js implementation of an MCP (Model Context Protocol) server with chat and consensus tools. Built with modern Node.js practices and official SDKs for seamless AI provider integration.
 
 ## 🚀 Quick Start
 
-### Option 1: Direct from GitHub (Recommended)
+### Option 1: Direct from NPM (Recommended)
 
 ```bash
 # Using npx (recommended)
-npx FallDownTheSystem/converse
+npx converse-mcp-server
 
 # Using pnpm dlx (alternative)
-pnpm dlx FallDownTheSystem/converse
+pnpm dlx converse-mcp-server
 
 # Using yarn dlx (alternative)  
-yarn dlx FallDownTheSystem/converse
+yarn dlx converse-mcp-server
 ```
 
 ### Option 2: Clone and Install
@@ -63,7 +65,6 @@ LOG_LEVEL=info
 MAX_MCP_OUTPUT_TOKENS=200000
 
 # Optional: Provider-specific settings
-GOOGLE_LOCATION=us-central1
 XAI_BASE_URL=https://api.x.ai/v1
 OPENROUTER_REFERER=https://github.com/FallDownTheSystem/converse
 ```
@@ -91,7 +92,7 @@ There are several ways to add the Converse MCP Server to Claude:
   "mcpServers": {
     "converse": {
       "command": "npx",
-      "args": ["FallDownTheSystem/converse"],
+      "args": ["converse-mcp-server"],
       "env": {
         "OPENAI_API_KEY": "your_key_here",
         "GOOGLE_API_KEY": "your_key_here",
@@ -108,31 +109,7 @@ There are several ways to add the Converse MCP Server to Claude:
 }
 ```
 
-#### Option B: Using NPX with stdio transport
-
-```json
-{
-  "mcpServers": {
-    "converse": {
-      "command": "npx",
-      "args": ["FallDownTheSystem/converse", "--transport", "stdio"],
-      "env": {
-        "OPENAI_API_KEY": "your_key_here",
-        "GOOGLE_API_KEY": "your_key_here",
-        "XAI_API_KEY": "your_key_here",
-        "ANTHROPIC_API_KEY": "your_key_here",
-        "MISTRAL_API_KEY": "your_key_here",
-        "DEEPSEEK_API_KEY": "your_key_here",
-        "OPENROUTER_API_KEY": "your_key_here",
-        "OPENROUTER_REFERER": "https://github.com/YourUsername/YourApp",
-        "MAX_MCP_OUTPUT_TOKENS": "200000"
-      }
-    }
-  }
-}
-```
-
-#### Option C: Direct Node.js execution
+#### Option B: Direct Node.js execution
 
 ```json
 {
@@ -140,9 +117,7 @@ There are several ways to add the Converse MCP Server to Claude:
     "converse": {
       "command": "node",
       "args": [
-        "C:\\Users\\YourUsername\\Documents\\Projects\\converse\\src\\index.js",
-        "--transport",
-        "stdio"
+        "C:\\Users\\YourUsername\\Documents\\Projects\\converse\\src\\index.js"
       ],
       "env": {
         "OPENAI_API_KEY": "your_key_here",
@@ -160,27 +135,7 @@ There are several ways to add the Converse MCP Server to Claude:
 }
 ```
 
-#### Option D: Using environment variable for transport
-
-```json
-{
-  "mcpServers": {
-    "converse": {
-      "command": "npx",
-      "args": ["FallDownTheSystem/converse"],
-      "env": {
-        "MCP_TRANSPORT": "stdio",
-        "OPENAI_API_KEY": "your_key_here",
-        "GOOGLE_API_KEY": "your_key_here",
-        "XAI_API_KEY": "your_key_here",
-        "MAX_MCP_OUTPUT_TOKENS": "200000"
-      }
-    }
-  }
-}
-```
-
-#### Option E: Local HTTP Development (Advanced)
+#### Option C: Local HTTP Development (Advanced)
 
 For local development with HTTP transport (optional, for debugging):
 
@@ -207,9 +162,16 @@ For local development with HTTP transport (optional, for debugging):
 #### Installation Steps
 
 1. **For Claude Code**: 
-   - Open the command palette (Ctrl/Cmd + Shift + P)
-   - Run "Claude Code: Edit MCP Settings"
-   - Add one of the configurations above
+   ```bash
+   # Add the server globally (for all projects)
+   claude mcp add converse npx converse-mcp-server -s user
+   
+   # Then set your API keys
+   claude mcp set-env converse OPENAI_API_KEY=your_key_here -s user
+   claude mcp set-env converse GOOGLE_API_KEY=your_key_here -s user
+   claude mcp set-env converse XAI_API_KEY=your_key_here -s user
+   # Add other API keys as needed
+   ```
 
 2. **For Claude Desktop**:
    - Navigate to Settings → Developer → MCP Servers
@@ -291,7 +253,9 @@ Programmatic access to documentation:
 ### OpenAI Models
 - **o3**: Strong reasoning (200K context)
 - **o3-mini**: Fast O3 variant (200K context)  
+- **o3-pro**: Professional-grade reasoning (200K context) - EXTREMELY EXPENSIVE
 - **o4-mini**: Latest reasoning model (200K context)
+- **gpt-4.1**: Advanced reasoning (1M context)
 - **gpt-4o**: Multimodal flagship (128K context)
 - **gpt-4o-mini**: Fast multimodal (128K context)
 
@@ -299,12 +263,34 @@ Programmatic access to documentation:
 - **gemini-2.5-flash** (alias: `flash`): Ultra-fast (1M context)
 - **gemini-2.5-pro** (alias: `pro`): Deep reasoning (1M context)
 - **gemini-2.0-flash**: Latest with experimental thinking
+- **gemini-2.0-flash-lite**: Lightweight fast model, text-only
 
 ### X.AI/Grok Models  
 - **grok-4-0709** (alias: `grok`): Latest advanced model (256K context)
 - **grok-3**: Previous generation (131K context)
 - **grok-3-fast**: Higher performance variant
 
+### Anthropic Models
+- **claude-opus-4**: Highest intelligence with extended thinking (200K context)
+- **claude-sonnet-4**: Balanced performance with extended thinking (200K context)
+- **claude-3.7-sonnet**: Enhanced 3.x generation with thinking (200K context)
+- **claude-3.5-sonnet**: Fast and intelligent (200K context)
+- **claude-3.5-haiku**: Fastest model for simple queries (200K context)
+
+### Mistral Models
+- **magistral-medium**: Frontier-class reasoning model (40K context)
+- **magistral-small**: Small reasoning model (40K context)
+- **mistral-medium-3**: Frontier-class multimodal model (128K context)
+
+### DeepSeek Models
+- **deepseek-chat**: Strong MoE model with 671B/37B parameters (64K context)
+- **deepseek-reasoner**: Advanced reasoning model with CoT (64K context)
+
+### OpenRouter Models
+- **qwen3-235b-thinking**: Qwen3 with enhanced reasoning (32K context)
+- **qwen3-coder**: Specialized for programming tasks (32K context)
+- **kimi-k2**: Moonshot AI Kimi K2 with extended context (200K context)
+
 ## 🚀 Development
 
 ### Install from Source
@@ -342,8 +328,12 @@ npm run kill-server    # Kill any server running on port 3157
 npm test               # Run all tests
 npm run test:unit      # Unit tests only
 npm run test:integration # Integration tests
+npm run test:mcp-client # MCP client tests (HTTP-based client-server testing)
 npm run test:real-api  # Real API tests (requires keys)
+npm run test:providers # Provider tests
+npm run test:tools     # Tool tests
 npm run test:coverage  # Coverage report
+npm run test:watch     # Run tests in watch mode
 
 # Code quality
 npm run lint           # Check code style
@@ -396,7 +386,7 @@ XAI_API_KEY=xai-...
 npm run test:real-api
 
 # Run comprehensive integration tests
-node final-integration-test.js
+node tests/integration/final-integration-test.js
 
 # Validate server functionality
 npm run validate
@@ -417,14 +407,14 @@ npm test
 npm run test:real-api
 
 # 4. Comprehensive validation
-node final-integration-test.js
+node tests/integration/final-integration-test.js
 ```
 
 **Expected Results:**
 - Server starts without errors on port 3157
 - All unit tests pass
 - Real API tests connect successfully (if keys configured)
-- Integration tests achieve >70% success rate
+- Some real API integration tests may occasionally timeout
 
 ## 📦 Publishing to NPM
 
@@ -530,7 +520,6 @@ converse/
 | `PORT` | Server port | `3157` | `3157` |
 | `LOG_LEVEL` | Logging level | `info` | `debug`, `info`, `error` |
 | `MAX_MCP_OUTPUT_TOKENS` | Token response limit | `25000` | `200000` |
-| `GOOGLE_LOCATION` | Google API region | `us-central1` | `us-central1` |
 | `XAI_BASE_URL` | XAI API endpoint | `https://api.x.ai/v1` | Custom endpoint |
 
 ### Model Selection
@@ -633,7 +622,7 @@ git push origin feature/your-feature
 
 ## 🙏 Acknowledgments
 
-This MCP Server was inspired by and builds upon the excellent work from [BeehiveInnovations/zen-mcp-server](https://github.com/BeehiveInnovations/zen-mcp-server). We're grateful for their pioneering implementation and innovative approach to MCP server development.
+This MCP Server was inspired by and builds upon the excellent work from [BeehiveInnovations/zen-mcp-server](https://github.com/BeehiveInnovations/zen-mcp-server).
 
 ## 📄 License
 
@@ -643,8 +632,4 @@ MIT License - see [LICENSE](LICENSE) file for details.
 
 - **GitHub**: https://github.com/FallDownTheSystem/converse
 - **Issues**: https://github.com/FallDownTheSystem/converse/issues
-- **NPM Package**: https://www.npmjs.com/package/converse-mcp-server
-
----
-
-**Built with ❤️ using Node.js and modern AI APIs**
+- **NPM Package**: https://www.npmjs.com/package/converse-mcp-server

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "converse-mcp-server",
-  "version": "1.5.1",
+  "version": "1.5.3",
   "description": "Converse MCP Server - Converse with other LLMs with chat and consensus tools",
   "type": "module",
   "main": "src/index.js",

package/src/providers/anthropic.js

@@ -351,19 +351,25 @@ export const anthropicProvider = {
     // Get Anthropic SDK
     const Anthropic = await getAnthropicSDK();
 
+    // Resolve model name first
+    const resolvedModel = resolveModelName(model);
+    const modelConfig = SUPPORTED_MODELS[resolvedModel] || {};
+
     // Initialize Anthropic client with default headers
     // Use both prompt caching and extended cache duration headers for 1-hour caching
+    // Add thinking beta for models that support thinking
+    const betaHeaders = ['prompt-caching-2024-07-31', 'extended-cache-ttl-2025-04-11'];
+    if (modelConfig.supportsThinking && reasoning_effort) {
+      betaHeaders.push('thinking-2025-01-27');
+    }
+    
     const anthropic = new Anthropic({
       apiKey: config.apiKeys.anthropic,
       defaultHeaders: {
-        'anthropic-beta': 'prompt-caching-2024-07-31,extended-cache-ttl-2025-04-11'
+        'anthropic-beta': betaHeaders.join(',')
       }
     });
 
-    // Resolve model name
-    const resolvedModel = resolveModelName(model);
-    const modelConfig = SUPPORTED_MODELS[resolvedModel] || {};
-
     // Convert messages to Anthropic format (system messages are always cached)
     const { systemPrompt, messages: anthropicMessages } = convertMessagesToAnthropic(messages);
 
@@ -380,39 +386,38 @@ export const anthropicProvider = {
       requestPayload.system = systemPrompt;
     }
 
-    // Add max tokens (required by Anthropic)
-    const defaultMaxTokens = modelConfig.maxOutputTokens || 8192;
-
-    // If thinking is supported and enabled, we need to reduce max_tokens to leave room for thinking
-    let effectiveMaxTokens = defaultMaxTokens;
-    if (modelConfig.supportsThinking && reasoning_effort) {
-      // Reserve some tokens for thinking - use a more conservative approach
-      effectiveMaxTokens = Math.min(defaultMaxTokens, 16000); // Cap at 16k for models with thinking
+    // Add max tokens only if explicitly requested
+    // For Claude 4 series models, let the SDK use its defaults (32k for opus, 64k for sonnet)
+    if (maxTokens) {
+      requestPayload.max_tokens = Math.min(maxTokens, modelConfig.maxOutputTokens || 8192);
+    } else if (!resolvedModel.includes('claude-opus-4') && !resolvedModel.includes('claude-sonnet-4')) {
+      // For non-4 series models, we still need to set max_tokens
+      requestPayload.max_tokens = modelConfig.maxOutputTokens || 8192;
     }
-
-    requestPayload.max_tokens = maxTokens
-      ? Math.min(maxTokens, effectiveMaxTokens)
-      : effectiveMaxTokens;
+    // For 4 series models without explicit maxTokens, don't set max_tokens - let SDK use defaults
 
     // Add thinking configuration for models that support it
     if (modelConfig.supportsThinking && reasoning_effort) {
       const thinkingBudget = calculateThinkingBudget(modelConfig, reasoning_effort);
-      if (thinkingBudget > 0) {
-        // Anthropic docs: thinking budget counts towards total token limit
-        // So we need to ensure max_tokens + budget_tokens <= model's actual limit
-        // Reduce max_tokens to make room for thinking
-        const reducedMaxTokens = requestPayload.max_tokens - thinkingBudget;
-
-        if (reducedMaxTokens >= 1000 && thinkingBudget >= 1024) { // Ensure we have reasonable space for both
-          requestPayload.max_tokens = reducedMaxTokens;
-          requestPayload.thinking = {
-            type: 'enabled',
-            budget_tokens: thinkingBudget
-          };
-          debugLog(`[Anthropic] Thinking enabled with budget: ${thinkingBudget} tokens, max_tokens reduced to: ${reducedMaxTokens} (${reasoning_effort} effort)`);
-        } else {
-          debugLog(`[Anthropic] Not enough token budget for thinking. Would need ${thinkingBudget} thinking + ${reducedMaxTokens} output tokens`);
-        }
+      debugLog(`[Anthropic] Model ${resolvedModel}: maxOutputTokens=${modelConfig.maxOutputTokens}, maxThinkingTokens=${modelConfig.maxThinkingTokens}, thinkingBudget=${thinkingBudget}`);
+      
+      // For 4 series models, we trust the SDK defaults work with thinking
+      // For other models, check against max_tokens if set
+      const maxTokensLimit = requestPayload.max_tokens || 
+        (resolvedModel.includes('claude-opus-4') ? 32000 : 
+         resolvedModel.includes('claude-sonnet-4') ? 64000 : 
+         modelConfig.maxOutputTokens);
+      
+      if (thinkingBudget > 0 && thinkingBudget < maxTokensLimit) {
+        // According to Anthropic docs: thinking tokens count towards max_tokens limit
+        // thinking.budget_tokens must be >= 1024 and < max_tokens
+        requestPayload.thinking = {
+          type: 'enabled',
+          budget_tokens: thinkingBudget
+        };
+        debugLog(`[Anthropic] Thinking enabled with budget: ${thinkingBudget} tokens (${reasoning_effort} effort)`);
+      } else {
+        debugLog(`[Anthropic] Thinking not enabled: budget ${thinkingBudget} must be < max_tokens limit ${maxTokensLimit}`);
       }
     }
 
@@ -429,6 +434,14 @@ export const anthropicProvider = {
 
     try {
       debugLog(`[Anthropic] Calling ${resolvedModel} with ${anthropicMessages.length} messages`);
+      debugLog(`[Anthropic] Request payload:`, JSON.stringify({
+        model: requestPayload.model,
+        max_tokens: requestPayload.max_tokens,
+        thinking: requestPayload.thinking,
+        temperature: requestPayload.temperature,
+        message_count: requestPayload.messages?.length,
+        system_length: Array.isArray(requestPayload.system) ? requestPayload.system[0]?.text?.length : requestPayload.system?.length
+      }, null, 2));
       if (systemPrompt) {
         debugLog(`[Anthropic] System prompt length: ${systemPrompt.length} characters`);
       }
@@ -507,8 +520,21 @@ export const anthropicProvider = {
         throw new AnthropicProviderError(`Invalid request: ${error.error.message}`, ErrorCodes.INVALID_REQUEST, error);
       } else if (error.error?.type === 'not_found_error') {
         throw new AnthropicProviderError(`Model ${resolvedModel} not found`, ErrorCodes.MODEL_NOT_FOUND, error);
-      } else if (error.message?.includes('context length') || error.message?.includes('token')) {
-        throw new AnthropicProviderError('Context length exceeded for model', ErrorCodes.CONTEXT_LENGTH_EXCEEDED, error);
+      } else if (error.message?.includes('context length') || error.message?.includes('context_length') || 
+                 (error.message?.includes('token') && error.message?.includes('limit'))) {
+        debugError(`[Anthropic] Context length error - Full error:`, error);
+        debugError(`[Anthropic] Error message:`, error.message);
+        debugError(`[Anthropic] Error response:`, error.response);
+        throw new AnthropicProviderError(`Context length exceeded for model: ${error.message}`, ErrorCodes.CONTEXT_LENGTH_EXCEEDED, error);
+      } else if (error.message?.includes('Streaming is strongly recommended')) {
+        // This is just a warning from the SDK about long requests
+        debugLog(`[Anthropic] SDK streaming recommendation warning`);
+        debugError(`[Anthropic] Full error object:`, error);
+        // Check if there's an actual error response
+        if (error.response || error.status) {
+          debugError(`[Anthropic] Error response status:`, error.status);
+          debugError(`[Anthropic] Error response data:`, error.response);
+        }
       }
 
       // Generic error handling

package/src/providers/google.js

@@ -18,11 +18,11 @@ const SUPPORTED_MODELS = {
     supportsStreaming: true,
     supportsImages: true,
     supportsTemperature: true,
-    supportsThinking: true,
+    supportsThinking: false,
     supportsWebSearch: true,
-    maxThinkingTokens: 24576,
+    maxThinkingTokens: 0,
     timeout: 300000,
-    description: 'Gemini 2.0 Flash (1M context) - Latest fast model with experimental thinking, supports audio/video input and grounding',
+    description: 'Gemini 2.0 Flash (1M context) - Latest fast model, supports audio/video input and grounding',
     aliases: ['flash-2.0', 'flash2', 'flash 2.0', 'gemini flash 2.0', 'gemini-2.0-flash-latest']
   },
   'gemini-2.0-flash-lite': {