server-anthropic

Anthropic API MCP Server

A Model Context Protocol (MCP) server that provides access to Anthropic's AI models through their official API. List available models and send messages to Claude using a secure, standardized interface. More about MCP.

Features

• List all available Anthropic models
• Send messages to any Anthropic model
• Secure API key management
• Support for workspace segmentation
• Automatic API versioning
• Comprehensive error handling
• Experimental context window management

Why Use This Server?

Unlike direct API integration, this server:

• Provides a standardized MCP interface for Anthropic's API
• Handles authentication and versioning automatically
• Supports workspace isolation for different use cases
• Integrates seamlessly with Claude Desktop and other MCP clients
• Includes detailed error messages and validation
• Potentially extends conversation context windows through distributed processing

Context Window Management Hypothesis

This server explores efficient context window management through distributed processing:

Claim: Offloading processing to separate API calls through the MCP server may reduce context window pressure on the main Claude Desktop conversation.

Primary Hypothesis: By branching conversation processing across multiple isolated API instances, the total effective context window of a conversation can be extended beyond standard limitations.

Key Mechanisms:

• Main conversation (trunk) retains only final outputs and discussion
• Individual API calls maintain isolated contexts
• Processing can be distributed across multiple model instances

Testing Scenarios:

• Context Window Extension

  • Compare context limits with and without API offloading
  • Measure conversation longevity in both scenarios
  • Track rate limit encounters

• Processing Distribution

  • Haiku for rapid initial processing
  • Sonnet for medium-sized aggregation
  • Opus for comprehensive analysis

Installation

npm install server-anthropic

Tool Reference

list_models

Lists all available Anthropic models and their capabilities.

Arguments:

{
  // No arguments required
}

Returns:

{
  "models": [
    {
      "name": "claude-3-opus-20240229",
      "description": "Most powerful model for highly complex tasks",
      ...
    },
    ...
  ]
}

send_message

Send a message to an Anthropic model using the Messages API.

Arguments:

{
  "messages": {
    "type": "array",
    "description": "Array of messages to send",
    "items": {
      "role": "user | assistant",
      "content": "string"
    },
    "required": true
  },
  "model": {
    "type": "string",
    "description": "Model ID to use",
    "default": "claude-3-opus-20240229"
  },
  "max_tokens": {
    "type": "number",
    "description": "Maximum tokens to generate",
    "default": 1024
  }
}

Returns:

{
  "content": [
    {
      "type": "text",
      "text": "Model response..."
    }
  ]
}

Usage with Claude Desktop

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "anthropic": {
      "command": "npx",
      "args": ["-y", "server-anthropic"],
      "env": {
        "ANTHROPIC_API_KEY": "your-api-key-here",
        "ANTHROPIC_WORKSPACE_ID": "optional-workspace-id"
      }
    }
  }
}

Environment Variables

Variable	Required	Description
ANTHROPIC_API_KEY	Yes	Your Anthropic API key from console.anthropic.com
ANTHROPIC_WORKSPACE_ID	No	Optional workspace ID for usage segmentation

Dependencies

• @modelcontextprotocol/sdk - Core MCP functionality
• @anthropic-ai/sdk - Official Anthropic API client

Error Handling

The server provides detailed error messages for common issues:

• Invalid API key
• Missing required parameters
• Rate limiting
• Model-specific errors
• Network connectivity issues

Development

• Clone the repository
• Install dependencies: npm install
• Set up environment variables in .env:

  ANTHROPIC_API_KEY=your-api-key
  ANTHROPIC_WORKSPACE_ID=optional-workspace
  

• Start the server: npm start

Version Notes (0.1.0-beta)

Current beta version focuses on:

• Core MCP server functionality
• Context window management testing
• API integration stability
• Initial hypothesis validation

Known limitations:

• Context window behavior requires further testing
• Rate limiting patterns under investigation
• Workspace segmentation impact on context management undefined

License

MIT