mcp-chat-analysis

The MCP Chat Analysis Server provides semantic understanding and exploration of chat conversations. It leverages a vector search backend (Qdrant) for semantic similarity, a knowledge graph (Neo4j) to model relationships between messages, concepts, and topics, and analytics capabilities to measure conversation dynamics. Typical workflows include importing conversations, performing semantic searches to locate relevant messages, extracting core concepts, and analyzing metrics such as message frequency and response times. The server is designed to integrate with MCP-compatible tooling (e.g., Claude Desktop) to enable seamless in-context analysis and retrieval within larger workflows.

To use the available tools, start the server with the MCP entry point and configure your databases via the provided environment variables. The available tools include import_conversations to ingest chat exports, semantic_search to find conversations by meaning rather than exact keywords, analyze_metrics to compute metrics around a conversation, and extract_concepts to pull out key ideas and topics. Each tool accepts a JSON payload describing its parameters, such as source paths for imports, query terms for searches, and identifiers for specific conversations. This setup enables end-to-end analysis from ingestion to insight, all accessible through MCP-compatible interfaces.

Prerequisites

• Python 3.8+
• Docker (optional for database services like Neo4j and Qdrant)
• Access to a Unix-like environment (Linux/macOS) or Windows with WSL2

Installation steps

1. Install the package

pip install mcp-chat-analysis-server

2. Set up databases (recommended Docker-based setup)

# Start containers for Qdrant and Neo4j (adjust versions as needed)
docker compose up -d

3. Configure the server

cp .env.example .env
# Edit .env with your settings (e.g., QDRANT_URL, NEO4J_URL, credentials)

4. Run the server

python -m mcp_chat_analysis.server

Development and testing

git clone https://github.com/rebots-online/mcp-chat-analysis-server.git
cd mcp-chat-analysis-server
pytest tests/

Environment variables:

• QDRANT_URL: URL of the Qdrant vector database used for semantic search.
• NEO4J_URL, NEO4J_USER, NEO4J_PASSWORD: Credentials and URL for the Neo4j knowledge graph. Common issues:
• Connection errors to Qdrant or Neo4j typically indicate network access problems or misconfigured URLs/credentials. Verify that the services are running and reachable from the host running the MCP server.
• If importing large conversation datasets, ensure sufficient memory and adjust batch sizes in the import tool as needed.
• When using Docker-based deployments for databases, ensure the containers are up and the correct ports are exposed to the MCP server. Configuration tips:
• Keep sensitive credentials in a secure .env file and avoid committing it to version control.
• Use the MCP integration block in claude_desktop_config.json to connect this server with Claude or other MCP clients.