---

Overview

A powerful MCP server that provides intelligent summarization capabilities through a clean, extensible architecture. Built with modern TypeScript and designed for seamless integration with AI workflows.

Installation

Installing via Smithery

To install Summarization Functions for Claude Desktop automatically via Smithery:

npx -y @smithery/cli install mcp-summarization-functions --client claude

npm i mcp-summarization-functions

AI Agent Integration

This MCP server was primarily developed to enhance the performance and reliability of AI agents like Roo Cline and Cline. It addresses a critical challenge in AI agent operations: context window management.

Context Window Optimization

AI agents frequently encounter situations where their context window gets rapidly filled with large outputs from:

• Command execution results
• File content readings
• Directory listings
• API responses
• Error messages and stack traces

This server helps maintain efficient context usage by:

1. Providing concise, relevant summaries instead of full content
2. Storing full content for reference when needed
3. Offering focused analysis based on specific needs (security, API surface, etc.)
4. Supporting multiple output formats for optimal context utilization

Benefits for AI Agents

• Reduced Failure Rates: By preventing context window overflow
• Improved Response Quality: Through focused, relevant summaries
• Enhanced Efficiency: By maintaining important context while reducing noise
• Better Resource Management: Through intelligent content caching and retrieval
• Flexible Integration: Supporting multiple AI providers and configuration options

Recommended AI Agent Prompt

When integrating with AI agents, include the following in your agent's instructions:

# CONTEXT MANAGEMENT

You have access to summarization functions through the MCP server. These functions are NOT optional - you MUST use them for ALL potentially large outputs to prevent context overflow:

MANDATORY SUMMARIZATION:
- You MUST ALWAYS use summarization functions for:
    - ANY first time file reading operations (unless you are CERTAIN its small and you are going to edit it)
    - ALL command execution outputs
    - EVERY directory analysis
    - ANY API responses or error logs
    - ANY output that could be large

NEVER attempt to process raw output directly - ALWAYS use the appropriate summarization function:
• For commands: summarize_command
• For files: summarize_files
• For directories: summarize_directory
• For other text: summarize_text

ALWAYS utilize available features:
• Specify hints for focused analysis
• Choose appropriate output formats
• Use content IDs to access full details only when absolutely necessary

There is NO NEED to process perfect or complete output. Summarized content is ALWAYS preferred over raw data. When in doubt, use summarization.

Summarization in action on the Ollama repository (Gemini 2.0 Flash summarization, Claude 3.5 agent)

Features

• Command Output Summarization
  Execute commands and get concise summaries of their output

• File Content Analysis
  Summarize single or multiple files while maintaining technical accuracy

• Directory Structure Understanding
  Get clear overviews of complex directory structures

• Flexible Model Support Use models from different providers

• AI Agent Context Optimization Prevent context window overflow and improve AI agent performance through intelligent summarization

Configuration

The server supports multiple AI providers through environment variables:

Required Environment Variables

• PROVIDER: AI provider to use. Supported values: - ANTHROPIC - Claude models from Anthropic - OPENAI - GPT models from OpenAI - OPENAI-COMPATIBLE - OpenAI-compatible APIs (e.g. Azure) - GOOGLE - Gemini models from Google
• API_KEY: API key for the selected provider

Optional Environment Variables

• MODEL_ID: Specific model to use (defaults to provider's standard model)
• PROVIDER_BASE_URL: Custom API endpoint for OpenAI-compatible providers
• MAX_TOKENS: Maximum tokens for model responses (default: 1024)
• SUMMARIZATION_CHAR_THRESHOLD: Character count threshold for when to summarize (default: 512)
• SUMMARIZATION_CACHE_MAX_AGE: Cache duration in milliseconds (default: 3600000 - 1 hour)
• MCP_WORKING_DIR - fallback directory for trying to find files with relative paths from

Example Configurations

# Anthropic Configuration
PROVIDER=ANTHROPIC
API_KEY=your-anthropic-key
MODEL_ID=claude-3-5-sonnet-20241022

# OpenAI Configuration
PROVIDER=OPENAI
API_KEY=your-openai-key
MODEL_ID=gpt-4-turbo-preview

# Azure OpenAI Configuration
PROVIDER=OPENAI-COMPATIBLE
API_KEY=your-azure-key
PROVIDER_BASE_URL=https://your-resource.openai.azure.com
MODEL_ID=your-deployment-name

# Google Configuration
PROVIDER=GOOGLE
API_KEY=your-google-key
MODEL_ID=gemini-2.0-flash-exp

Usage

Add the server to your MCP configuration file:

{
		"mcpServers": {
				"MUST_USE_summarization": {
						"command": "node",
						"args": ["path/to/summarization-functions/build/index.js"],
						"env": {
								"PROVIDER": "ANTHROPIC",
								"API_KEY": "your-api-key",
								"MODEL_ID": "claude-3-5-sonnet-20241022",
                "MCP_WORKING_DIR": "default_working_directory"
						}
				}
		}
}

Available Functions

The server provides the following summarization tools:

summarize_command

Execute and summarize command output.

{
  // Required
  command: string,    // Command to execute
  cwd: string,       // Working directory for command execution
  
  // Optional
  hint?: string,      // Focus area: "security_analysis" | "api_surface" | "error_handling" | "dependencies" | "type_definitions"
  output_format?: string  // Format: "text" | "json" | "markdown" | "outline" (default: "text")
}

summarize_files

Summarize file contents.

{
  // Required
  paths: string[],    // Array of file paths to summarize (relative to cwd)
  cwd: string,       // Working directory for resolving file paths
  
  // Optional
  hint?: string,      // Focus area: "security_analysis" | "api_surface" | "error_handling" | "dependencies" | "type_definitions"
  output_format?: string  // Format: "text" | "json" | "markdown" | "outline" (default: "text")
}

summarize_directory

Get directory structure overview.

{
  // Required
  path: string,       // Directory path to summarize (relative to cwd)
  cwd: string,       // Working directory for resolving directory path
  
  // Optional
  recursive?: boolean,  // Whether to include subdirectories. Safe for deep directories
  hint?: string,       // Focus area: "security_analysis" | "api_surface" | "error_handling" | "dependencies" | "type_definitions"
  output_format?: string   // Format: "text" | "json" | "markdown" | "outline" (default: "text")
}

summarize_text

Summarize arbitrary text content.

{
  // Required
  content: string,    // Text content to summarize
  type: string,       // Type of content (e.g., "log output", "API response")
  
  // Optional
  hint?: string,      // Focus area: "security_analysis" | "api_surface" | "error_handling" | "dependencies" | "type_definitions"
  output_format?: string  // Format: "text" | "json" | "markdown" | "outline" (default: "text")
}

get_full_content

Retrieve the full content for a given summary ID.

{
  // Required
  id: string         // ID of the stored content
}

Running evals

The evals package loads an mcp client that then runs the index.ts file, so there is no need to rebuild between tests. You can load environment variables by prefixing the npx command. Full documentation can be found here.

OPENAI_API_KEY=your-key  npx mcp-eval src/evals/evals.ts src/server/mcp-server.ts

License

MIT