alonlevyshavit/ai-context-mcp
If you are the rightful owner of ai-context-mcp and would like to certify it and/or have it hosted online, please leave a comment on the right or send an email to dayong@mcphub.com.
The AI Context MCP Server is a standalone server that provides AI context orchestration for projects with a `.ai-context` folder, enabling dynamic tool generation and metadata extraction.
AI Context MCP Server
A standalone MCP (Model Context Protocol) server that provides AI context orchestration for any project with a .ai-context folder. This server runs independently and can be used by multiple projects without installation via npx.
Quick Start
IMPORTANT: The AI_CONTEXT_ROOT environment variable is required and must point to your .ai-context folder.
Cursor (.cursor/mcp.json):
{
"mcpServers": {
"ai-context": {
"command": "npx",
"args": ["--yes", "github:alonlevyshavit/ai-context-mcp"],
"env": {
"AI_CONTEXT_ROOT": "/absolute/path/to/your/project/.ai-context"
}
}
}
}
Claude Desktop:
{
"mcpServers": {
"ai-context": {
"command": "npx",
"args": ["--yes", "github:alonlevyshavit/ai-context-mcp"],
"env": {
"AI_CONTEXT_ROOT": "/absolute/path/to/your/project/.ai-context"
}
}
}
}
The server will:
- Download automatically from GitHub
- Build itself if needed
- Use your specified
.ai-contextfolder - Provide tools for loading agents, guidelines, and frameworks
How It Works
Discovery-First Approach
The server embeds system instructions that guide AI assistants to:
- Discover available resources first using
list_all_resources - Read tool descriptions to understand each resource's purpose
- Select appropriate tools based on the task at hand
- Load resources strategically - single or multiple as needed
This ensures AI assistants adapt to your specific .ai-context structure rather than assuming certain agents exist.
Dynamic Tool Generation
The server automatically scans your .ai-context folder and creates specific MCP tools for each resource:
.ai-context/agents/planner.md → load_planner_agent tool
.ai-context/guidelines/api-design.md → load_guideline_api_design tool
.ai-context/frameworks/memory/README.md → load_framework_memory tool
Each tool includes rich metadata extracted from the resource files to help AI assistants make informed decisions about when to use them.
Multi-Format Metadata Support
The server supports multiple metadata formats with graceful fallbacks:
YAML Frontmatter (Preferred):
```markdown
description: Planning specialist for task breakdown and project organization use_cases:
- Breaking down complex tasks into smaller steps
- Creating project roadmaps
- Organizing work into manageable chunks
Planner Agent
[content...]
**HTML Comments:**
```markdown
<!-- metadata
description: Planning specialist for task breakdown and project organization
use_cases: Task breakdown; Project roadmaps; Work organization
-->
# Planner Agent
[content...]
Natural Language Fallback:
# Planner Agent
You are a planning specialist focused on breaking down complex tasks into manageable steps. You excel at creating structured roadmaps and organizing work efficiently.
[content...]
Configuration for Cursor
Create a .cursor/mcp.json file in your project root:
{
"mcpServers": {
"ai-context": {
"command": "npx",
"args": ["--yes", "github:alonlevyshavit/ai-context-mcp"],
"env": {
"AI_CONTEXT_ROOT": "/absolute/path/to/your/project/.ai-context"
}
}
}
}
Note: Replace /absolute/path/to/your/project/.ai-context with the actual absolute path to your .ai-context folder.
Required Configuration
The AI_CONTEXT_ROOT environment variable must be set to the absolute path of your .ai-context folder.
This explicit configuration ensures:
- Clear and predictable behavior
- No ambiguity about which
.ai-contextfolder is being used - Consistent operation across different environments
- Explicit control over the context being loaded
Project Structure
your-project/
├── .ai-context/ # Your AI context directory
│ ├── agents/ # Agent definitions (.md files)
│ ├── guidelines/ # Guidelines (.md files, can be nested)
│ └── frameworks/ # Framework folders with README.md
├── .cursor/
│ └── mcp.json # MCP server configuration
└── [your project files...]
Available Tools
The server provides both dynamic and static tools:
Dynamic Tools (Generated from Content)
load_[name]_agent- Load a specific agent (e.g.,load_debugger_agent,load_planner_agent)load_guideline_[path]- Load a specific guidelineload_framework_[name]- Load a framework's documentation
Static Tools (Always Available)
list_all_resources- Lists all discovered agents, guidelines, and frameworksload_multiple_resources- Load multiple resources simultaneously
Configuration Options
Required: AI_CONTEXT_ROOT Path
{
"mcpServers": {
"ai-context": {
"command": "npx",
"args": ["--yes", "github:alonlevyshavit/ai-context-mcp"],
"env": {
"AI_CONTEXT_ROOT": "/absolute/path/to/.ai-context"
}
}
}
}
Selective Resource Loading
Control which types of resources are loaded as tools by setting environment variables.
Default behavior:
- Agents: Always loaded (cannot be disabled)
- Guidelines: Disabled by default (opt-in with
AI_CONTEXT_LOAD_GUIDELINES="true") - Frameworks: Disabled by default (opt-in with
AI_CONTEXT_LOAD_FRAMEWORKS="true")
{
"mcpServers": {
"ai-context": {
"command": "npx",
"args": ["--yes", "github:alonlevyshavit/ai-context-mcp"],
"env": {
"AI_CONTEXT_ROOT": "/absolute/path/to/.ai-context"
// Agents are always loaded
// Guidelines and frameworks are disabled by default
}
}
}
}
Examples:
Load agents only (default behavior):
"env": {
"AI_CONTEXT_ROOT": "/absolute/path/to/.ai-context"
}
Load agents and guidelines:
"env": {
"AI_CONTEXT_ROOT": "/absolute/path/to/.ai-context",
"AI_CONTEXT_LOAD_GUIDELINES": "true"
}
Load all resources (agents, guidelines, and frameworks):
"env": {
"AI_CONTEXT_ROOT": "/absolute/path/to/.ai-context",
"AI_CONTEXT_LOAD_GUIDELINES": "true",
"AI_CONTEXT_LOAD_FRAMEWORKS": "true"
}
Specific Version/Branch
{
"mcpServers": {
"ai-context": {
"command": "npx",
"args": ["github:your-org/ai-context-mcp#v1.0.0"]
}
}
}
Development
Setup
git clone https://github.com/your-org/ai-context-mcp.git
cd ai-context-mcp
npm install
Commands
npm run build # Build TypeScript to JavaScript
npm run dev # Run in development mode
npm run test # Run tests
npm run test:coverage # Run tests with coverage
npm run typecheck # TypeScript type checking
npm run validate # Validate .ai-context metadata
Testing with Local Project
AI_CONTEXT_ROOT=/path/to/test-project/.ai-context npm run dev
Architecture
The server is built with:
- TypeScript with ES2022/Node16 modules
- MCP SDK for protocol communication
- Vitest for comprehensive testing (98%+ coverage)
- Dynamic tool generation from file system scanning
- Enum-based constants for maintainable code
- Comprehensive error handling and logging
Key components:
Scanner- Discovers and extracts metadata from resourcesLoader- Loads content and assembles contextsMetadataExtractor- Multi-format metadata extraction with fallbacksAiContextMCPServer- Main MCP server with dynamic tool generation
License
MIT - See LICENSE file for details
Contributing
- Fork the repository
- Create a feature branch
- Make your changes with tests
- Run
npm run test:run && npm run typecheck - Submit a pull request
The dist/ folder must be committed for npx github: compatibility.