README - Coding Conversations MCP Server by waleedf

Coding Conversations MCP Server

A Model Context Protocol (MCP) server that enables Claude to access external AI models for coding assistance through OpenRouter.

Features

Provides two MCP tools:
- call_external_model: Consult external AI models for coding assistance
- list_available_models: Discover available models and their capabilities
Integrates with OpenAI's GPT-4.1 models (nano, mini, and full versions)
Includes detailed model metadata (descriptions, use cases, strengths, cost tiers)
Provides cost estimation for API calls
Features enhanced error handling with detailed error messages

Installation

Clone the repository:

git clone https://github.com/your-username/coding-conversations-mcp.git
cd coding-conversations-mcp

Install dependencies:

npm install

Create your configuration file:

cp src/config/models-config.template.json src/config/models-config.json

Edit src/config/models-config.json and add your OpenRouter API key
Build the project:

npm run build

Configuration

Setting up models-config.json

Edit the src/config/models-config.json file to configure which models are available:

{
  "models": {
    "nano": {
      "endpoint": "https://openrouter.ai/api/v1/chat/completions",
      "modelId": "openai/gpt-4.1-nano",
      "token": "YOUR_OPENROUTER_API_KEY_HERE",
      "metadata": {
        "description": "GPT-4.1 Nano - Smallest and fastest GPT-4.1 variant",
        "useCase": "Quick code reviews, simple debugging tasks, and basic coding assistance",
        "strengths": ["Fast response time", "Cost-effective", "Good for simple tasks"],
        "costTier": "low"
      }
    },
    "mini": {
      "endpoint": "https://openrouter.ai/api/v1/chat/completions",
      "modelId": "openai/gpt-4.1-mini",
      "token": "YOUR_OPENROUTER_API_KEY_HERE",
      "metadata": {
        "description": "GPT-4.1 Mini - Mid-sized GPT-4.1 variant with good balance of speed and capability",
        "useCase": "Complex debugging, code optimization, and architectural suggestions",
        "strengths": ["Good balance of speed and capability", "Strong coding knowledge", "Detailed explanations"],
        "costTier": "medium"
      }
    },
    "full": {
      "endpoint": "https://openrouter.ai/api/v1/chat/completions",
      "modelId": "openai/gpt-4.1",
      "token": "YOUR_OPENROUTER_API_KEY_HERE",
      "metadata": {
        "description": "GPT-4.1 - Full-sized GPT-4.1 model with maximum capability",
        "useCase": "Complex system design, advanced debugging, and in-depth code analysis",
        "strengths": ["Most comprehensive understanding", "Best for complex problems", "Detailed and nuanced responses"],
        "costTier": "high"
      }
    }
  },
  "routing": {
    "default_model": "mini"
  }
}

You can add additional models by following the same pattern. The token field should contain your OpenRouter API key. The metadata field is optional but recommended for providing useful information about each model.

Configuring with Claude

For Claude VSCode Extension

Edit the settings file at: ~/Library/Application Support/Code/User/globalStorage/saoudrizwan.claude-dev/settings/cline_mcp_settings.json

Add your MCP server:

{
  "mcpServers": {
    "coding-conversations": {
      "command": "node",
      "args": ["/path/to/your/coding-conversations-mcp/dist/index.js"],
      "disabled": false,
      "autoApprove": []
    }
  }
}

For Claude Desktop App

Edit the settings file at: ~/Library/Application Support/Claude/claude_desktop_config.json

Add your MCP server similar to the VSCode example above.

Usage

Listing Available Models

To see what models are available and their capabilities, ask Claude to use the list_available_models tool:

Can you use the list_available_models tool to show me what AI models are available for coding assistance?

Claude will return a list of available models with their metadata, including descriptions, use cases, strengths, and cost tiers.

Consulting External Models

Once configured, ask Claude to use the call_external_model tool to consult an external model:

Can you use the call_external_model tool to ask the mini model about the best way to implement promise.all with a concurrency limit in JavaScript?

Claude will use the tool to communicate with the specified model through OpenRouter and present the response.

Tool Parameters

call_external_model

The call_external_model tool accepts the following parameters:

model (optional): The model to use (e.g., "nano", "mini", "full"). If not specified, the default model from the configuration will be used.
message (required): The question or problem you want to ask the external model.
context (optional): Additional context about the problem, such as code snippets, error messages, or relevant background information.

list_available_models

The list_available_models tool doesn't require any parameters.

Response Format

call_external_model

The tool returns a response with the following fields:

response: The text response from the external model.
modelUsed: The name of the model that was used.
modelInfo: Metadata about the model, including description, use case, strengths, and cost tier.
usage: Token usage information.
estimatedCost: Estimated cost of the API call (if available).

list_available_models

The tool returns a response with the following fields:

models: A list of available models with their metadata.
defaultModel: The name of the default model.

Development

src/index.ts - Main entry point and MCP server implementation
src/services/model-service.ts - Service handling communication with OpenRouter
src/types.ts - TypeScript types and interfaces
src/config/models-config.json - Configuration (not version controlled)

Error Handling

The server includes enhanced error handling with specific error messages for different types of errors:

Authentication errors (401/403)
Rate limiting errors (429)
Server errors (500+)
General API errors

This makes it easier to diagnose and fix issues when they occur.

Cost Estimation

The server provides an estimated cost for each API call based on the token usage and approximate pricing for each model. This helps track usage and estimate costs.

License

MIT