⚠️ DISCLAIMER: This MCP server is entirely written and maintained by AI (Claude/Gemini) with manual supervision. It is a personal test project created to explore AI-assisted development. Use at your own discretion.
An intelligent Model Context Protocol (MCP) server for saving, managing, and enhancing context with pre-processing strategies. This server helps you organize information efficiently by applying smart transformations like clarification, analysis, and search optimization.
-
Intelligent Context Storage: Save and organize contexts with metadata and tags
-
Pre-processing Strategies: Multiple configurable strategies to enhance context quality:
- Clarify: Improve content clarity by detecting and fixing ambiguous language
- Analyze: Comprehensive content analysis (word count, complexity, structure)
- Search: Extract keywords and enhance searchability
- Fetch: Detect and manage URLs and external data references
- Custom: Support for custom processing strategies
-
Context Models: Pre-configured models combining multiple strategies:
clarify: Focus on clarity improvementsearch_optimized: Optimize for searchabilityanalysis: Detailed content analysiscomprehensive: All strategies enabledweb_enhanced: For web content with URL handling
-
Context Management Tools:
- Save contexts with automatic or model-based processing
- Load contexts and discover related content
- List contexts with filtering by tags
- Delete contexts
- Query available models
npm install context-processornpm run buildnpm startOr in development mode:
npm run devCreate a context-models.json file in the project root to define custom models:
{
"models": [
{
"name": "my_model",
"description": "My custom context model",
"strategies": [
{
"name": "clarify",
"type": "clarify",
"enabled": true,
"config": {}
},
{
"name": "search",
"type": "search",
"enabled": true,
"config": {
"maxKeywords": 10
}
}
]
}
]
}Save content as context with optional pre-processing.
Parameters:
title(string, required): Title for the contextcontent(string, required): Content to savetags(string[], optional): Tags for organizing contextmetadata(object, optional): Additional metadatamodelName(string, optional): Context model to use for pre-processing
Example:
{
"title": "API Documentation",
"content": "This is an API with multiple endpoints...",
"tags": ["api", "documentation"],
"metadata": { "version": "1.0" },
"modelName": "comprehensive"
}Load a previously saved context and discover related contexts.
Parameters:
contextId(string, required): ID of the context to load
Response:
{
"context": { /* ContextItem */ },
"relatedContexts": [ /* ContextItem[] */ ]
}List all saved contexts with optional filtering.
Parameters:
tags(string[], optional): Filter by tagslimit(number, optional): Maximum number of contextsoffset(number, optional): Number of contexts to skip
List all available context models.
Response:
{
"models": [
{
"name": "clarify",
"description": "Model focused on clarifying content",
"strategyCount": 1
}
],
"total": 5
}Get detailed information about a specific model.
Parameters:
modelName(string, required): Name of the model
Delete a context by ID.
Parameters:
contextId(string, required): ID of the context to delete
Analyzes content for:
- Ambiguous pronouns (it, this, that)
- Passive voice usage
- Vague language (basically, kind of, sort of)
Provides a clarity score and suggestions for improvement.
- Extracts 10 most frequent meaningful keywords
- Filters out common stop words
- Recommends search queries for the content
Provides metrics:
- Word count and average word length
- Sentence and paragraph counts
- Content complexity assessment (low/medium/high)
- Detects URLs in content
- Identifies up to 5 external references
- Metadata about data sources
Contexts are stored as JSON files in the ./contexts directory. Each context file is named using its UUID:
contexts/
├── a1b2c3d4-e5f6-7g8h-9i0j-1k2l3m4n5o6p.json
├── b2c3d4e5-f6g7-h8i9-j0k1-l2m3n4o5p6q.json
└── ...
{
"title": "User Authentication Design",
"content": "The authentication system basically allows users to log in with their credentials. This approach is generally more secure than storing passwords in plain text. That said, the system needs better error handling.",
"tags": ["security", "authentication"],
"modelName": "comprehensive"
}This will:
- Clarify the vague language
- Analyze the content structure
- Extract key topics (authentication, security, passwords, etc.)
- Save all results to context storage
{
"contextId": "a1b2c3d4-e5f6-7g8h-9i0j-1k2l3m4n5o6p"
}Returns the saved context plus up to 5 related contexts that share tags.
ContextMCPServer
├── ContextStorage: File-based persistence
├── ContextPreprocessor: Strategy execution engine
└── MCP Protocol Handler: Tool definitions and execution
User Request
↓
MCP Server (Tool Handler)
↓
ContextPreprocessor (if model specified)
├─→ Strategy 1 (Clarify)
├─→ Strategy 2 (Analyze)
└─→ Strategy 3 (Search)
↓
ContextStorage (Save/Load)
↓
Response
All types are defined in src/types.ts:
ContextItem: Core context data structurePreProcessingStrategy: Strategy configurationContextModel: Model definition- Request/Response types for each tool
- Define the strategy type in
types.ts - Add a handler method in
ContextPreprocessor - Add the strategy to a model in
context-models.json
Example:
private customStrategy(
content: string,
config?: Record<string, unknown>
): PreProcessingResult {
// Your custom logic here
return {
strategy: "custom",
processed: true,
result: transformedContent,
};
}.
├── src/
│ ├── index.ts # Main MCP server
│ ├── types.ts # Type definitions
│ ├── storage.ts # Context persistence
│ └── preprocessor.ts # Processing strategies
├── contexts/ # Stored contexts (auto-created)
├── dist/ # Compiled output
├── context-models.json # Model configurations
├── package.json # Dependencies
└── tsconfig.json # TypeScript config
Run the built-in tests:
npm test- Database backend support (MongoDB, PostgreSQL)
- Vector embeddings for semantic search
- Machine learning-based categorization
- Multi-user context sharing
- Version control for contexts
- Integration with external APIs
- Real-time collaboration features
MIT