MCP Meeting Agent
A meeting preparation agent that provides trivia, fun facts, and GitHub trending repositories to improve your meetings. Built with LangChain's modern agent framework for intelligent orchestration and tool-based interactions.
Proof of Concept
This is a proof of concept application and is not intended for production use. This project demonstrates architectural patterns and integration concepts but lacks critical production features such as security, performance optimization, comprehensive testing, and scalability considerations.
Features
- Tech Trivia: Fetches technology trivia questions and answers from external APIs
- Fun Facts: Provides interesting random facts from external APIs
- GitHub Trending: Shows current trending repositories from external APIs
- Meeting Notes: Generates formatted meeting notes for hosts
- LangChain Agent Framework: Modern LLM agent architecture with tool-based orchestration
- Intelligent Tool Coordination: Agent automatically selects and uses appropriate tools
- Robust Error Handling: Graceful fallbacks and comprehensive error recovery
- Structured Logging: Comprehensive logging with configurable levels
- Comprehensive Testing: Full test coverage for all components
- FastMCP Integration: Production-ready MCP server with error masking, rate limiting, and context-aware logging
Architecture
The project follows a modern LangChain agent architecture with clean separation of concerns and provider-agnostic LLM support:
Provider-Agnostic Design
The application is designed to work with multiple LLM providers without code changes:
- OpenAI/OpenRouter: Use any OpenAI-compatible API via
base_urlconfiguration - Anthropic Claude: Direct support for Claude models (requires
langchain-anthropic) - Google Gemini: Direct support for Gemini models (requires
langchain-google-genai) - Other Providers: Any OpenAI-compatible API via
base_urlconfiguration
The system automatically detects the provider based on the model name and configuration, making it easy to switch between providers by simply updating your .env file.
- LangChain Agents: Intelligent orchestrators that coordinate tools using LLM reasoning
- Tools: Reusable LangChain tools that wrap external services and APIs
- Services: Handle external API interactions and data fetching
- Schemas: Pydantic models for data validation and structure
- Formatters: Format data for different output types (LLM, notes)
- Prompts: Manage LLM prompt templates
- Core: Configuration, logging, and LLM gateway
Key Architectural Benefits
- Tool-Based Design: Individual tools can be reused across different agents
- Intelligent Orchestration: Agent uses LLM reasoning to determine which tools to use
- Better Error Handling: Tools have individual error handling with graceful fallbacks
- Flexibility: Easy to add new tools without changing agent logic
- Modern LLM Patterns: Follows LangChain's recommended agent-tool architecture
Quick Start
-
Clone the repository:
git clone https://github.com/cliffordru/mcp-meeting-agent cd mcp-meeting-agent -
Install dependencies:
uv syncOptional: Install additional LLM provider support:
# For Anthropic Claude support uv add langchain-anthropic # For Google Gemini support uv add langchain-google-genai # Or install all providers uv add langchain-anthropic langchain-google-genai -
Set up environment variables:
cp env.example .env # Edit .env with your preferred settings -
Run the MCP Server:
uv run server.py -
Configure MCP Client: To use this MCP server with your MCP host client, see
mcp-client-config.jsonin the project root for the complete configuration example.The server runs on
http://127.0.0.1:8000/sseby default and provides theprepare_meeting()tool for generating meeting content. -
Use: For example in Cursor, you can prompt your AI assistant "Using your MCP tools, prepare some fun meeting notes".
Configuration
Key configuration options in .env:
# LLM Configuration (Provider Agnostic)
LLM_API_KEY=your_api_key
LLM_API_BASE_URL=https://api.openai.com/v1 # Optional: for OpenAI/OpenRouter
LLM_MODEL=gpt-4o-mini # Supports OpenAI, Anthropic Claude, Google Gemini
LLM_TEMPERATURE=0.0
LLM_REQUEST_TIMEOUT=60
# Provider Examples:
# OpenAI/OpenRouter: LLM_MODEL=gpt-4o-mini, LLM_API_BASE_URL=https://api.openai.com/v1
# Anthropic Claude: LLM_MODEL=claude-3-5-sonnet-20241022 (no base_url needed)
# Google Gemini: LLM_MODEL=gemini-1.5-flash (no base_url needed)
# API Configuration
TECH_TRIVIA_API_URL=your_tech_trivia_api_url
FUN_FACTS_API_URL=your_fun_facts_api_url
GITHUB_TRENDING_URL=your_github_trending_api_url
# Logging
LOG_LEVEL=INFO
# FastMCP Configuration
MCP_MASK_ERROR_DETAILS=true
MCP_ENABLE_LOGGING=true
Testing
Run all tests:
uv run pytest src/tests/ -v
Run with coverage:
uv run pytest src/tests/ --cov=src/app --cov-report=html
Production Readiness Considerations
Testing & Quality Assurance
Current State: Comprehensive unit tests with good coverage Production Needs:
- Load Testing: Apache Bench, Artillery, or k6 for API performance testing
- Integration Testing: End-to-end testing with real API dependencies
- Failure Scenario Testing: Network failures, API rate limits, LLM timeouts
- Evaluation Framework: TBD
- Monitoring: Langfuse integration for LLM performance tracking
Security Considerations
Current State: Basic FastMCP error masking and rate limiting implemented Production Needs:
- Authentication: OAuth 2.0 flow with proper session management
- Input Validation: Content filtering and sanitization for all inputs
- Output Filtering: LLM output validation to prevent harms
- SAST/DAST: Static and dynamic application security testing
- SCA: Software composition analysis for dependency vulnerabilities
- Rate Limiting: More sophisticated rate limiting and abuse prevention
- Secrets Management: TBD when needed
Performance & Scalability
Current State: Async implementation with LangChain agent framework Production Needs:
- Caching: TBD based on load testing
- Circuit Breakers: Resilience patterns for external API failures
- Horizontal Scaling: Container orchestration (Kubernetes/Docker)
AI Architecture Improvements
Current State: LangChain agent framework with tool-based architecture, LLM-generated fallback content for API failures, and provider-agnostic LLM support Production Needs:
- Model-as-a-Service: Right-sized models for cost/latency/accuracy balance
- Prompt Engineering: Systematic prompt optimization and versioning
- Agent Optimization: Fine-tune agent prompts and tool selection logic
- Multi-Provider Support: Provider switching and fallback mechanisms
- Tool Validation: Input/output validation for tools
- LLM-Generated Fallbacks: Improve dynamic LLM-generated content when APIs fail
- Generate contextual trivia questions based on meeting type/context
- Create relevant fun facts tailored to the audience/industry
- Provide trending tech topics specific to the team's domain
- Maintain content freshness and relevance through AI generation
Integration & Real-World Services
Current State: Basic external APIs with tool-based access Production Needs:
- GitHub Integration: Real-time issues, PRs, and repository health
- CI/CD Integration: Build status and deployment information
- Jira/Linear: Project management and sprint data
- Slack/Teams: Real-time notifications and team collaboration
- Calendar Integration: Meeting scheduling and participant management
- Analytics: Meeting effectiveness tracking and insights
Monitoring & Alerting
Current State: Basic structured logging with FastMCP client logging integration Production Needs:
- Observability & Alerting: Monitoring for agent performance and tool usage
- Centralized Logging: Log aggregation and analysis
- Performance Metrics: Response time tracking and alerting
Project Structure
mcp-meeting-agent/
├── src/
│ ├── app/
│ │ ├── agents/
│ │ │ ├── tech_trivia_agent.py
│ │ │ ├── fun_facts_agent.py
│ │ │ ├── github_trending_agent.py
│ │ │ └── meeting_planner_agent.py
│ │ ├── core/
│ │ │ ├── config.py
│ │ │ ├── llm_gateway.py
│ │ │ └── logging_config.py
│ │ ├── formatters/
│ │ │ ├── meeting_notes_formatter.py
│ │ │ └── repository_formatter.py
│ │ ├── prompts/
│ │ │ ├── __init__.py
│ │ │ ├── agent_prompts.py
│ │ │ └── fallback_prompts.py
│ │ ├── schemas/
│ │ │ ├── fun_facts.py
│ │ │ └── tech_trivia.py
│ │ ├── services/
│ │ │ ├── fun_facts_service.py
│ │ │ ├── github_trending_service.py
│ │ │ └── tech_trivia_service.py
│ │ └── tools/
│ │ └── meeting_tools.py
│ └── tests/
│ ├── test_fun_facts.py
│ ├── test_github_trending_agent.py
│ ├── test_github_trending_service.py
│ ├── test_meeting_notes.py
│ ├── test_meeting_planner_agent.py
│ ├── test_meeting_tools.py
│ ├── test_repository_formatter.py
│ ├── test_server_integration.py
│ └── test_tech_trivia_agent.py
├── server.py
├── env.example
├── .env (not tracked in git)
├── LICENSE
└── README.md
API Endpoints
The MCP server exposes a single tool:
prepare_meeting(ctx: Context, meeting_context: str = ""): Generates meeting preparation content including trivia, fun facts, and trending repositories using LangChain agent orchestration with error handling and context-aware logging
Dependencies
- FastMCP: MCP server framework
- aiohttp: Async HTTP client
- pydantic: Data validation
- structlog: Structured logging
- langchain: LLM integration and agent framework
- langchain-core: Core LangChain components
- langchain-openai: OpenAI integration for LangChain
- pytest: Testing framework
- pytest-asyncio: Async test support
Note: Langfuse observability is included in dependencies but not yet implemented in the current version.
License
This project is licensed under the MIT License - see the LICENSE file for details.