Context DB MCP (BETA Mode- can make mistakes)

A Model Context Protocol (MCP) server that stores and retrieves project context using OpenAI vector stores. Enable your AI editor to remember and recall project knowledge across sessions. For best results, add this in your agent rules file and force it to use it at the start and the very end of each chat session. Otherwise, in your prompt, ask the agent to use the ingestion or retrieval where necessary.

Features

• `ingest_document` - Store project summaries, design docs, and context in a vector store
• `retrieve_relevant_chunks` - Retrieve relevant context using semantic search

Installation

pip install -e .

Requirements: Python 3.11+ and an OpenAI API key

Configuration

Create a .env file (or set environment variables):

# Required
OPENAI_API_KEY=sk-your-api-key-here

# Optional - Vector Store (use ID for existing, NAME to create/find)
CONTEXT_DB_VECTOR_STORE_ID=vs_xxxxx
# OR
CONTEXT_DB_VECTOR_STORE_NAME=context-db-mcp

# Optional - Tuning
CONTEXT_DB_DEFAULT_MAX_RESULTS=10
CONTEXT_DB_REQUEST_TIMEOUT_SECONDS=120.0
CONTEXT_DB_LOG_LEVEL=INFO

Use env.example as a template.

Usage

Claude Code

Add to your project root inside .mcp.json (create with exact name and preceeding dot) and paste the following:

{
  "mcpServers": {
    "context-db": {
      "command": "{PATH TO THE BIN FOLDER IN YOUR CLONED REPO for example: {YOUR PATH.....}/Context_DB_MCP/env/bin/context-db-mcp}",
      "env": {
        "OPENAI_API_KEY": "sk-your-api-key-here",
        "CONTEXT_DB_VECTOR_STORE_NAME": "context-db-mcp"
      }
    }
  }
}

Restart Claude Code. Tools will be available as:

• mcp__context-db__ingest_document
• mcp__context-db__retrieve_relevant_chunks

Cursor

Add to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "context-db": {
      "command": "{PATH TO THE BIN FOLDER IN YOUR CLONED REPO for example: {YOUR PATH.....}/Context_DB_MCP/env/bin/context-db-mcp}",
      "env": {
        "OPENAI_API_KEY": "sk-your-api-key-here",
        "CONTEXT_DB_VECTOR_STORE_NAME": "context-db-mcp"
      }
    }
  }
}

Restart Cursor and access the tools from the MCP integration panel.

Testing

Run the diagnostic script to verify your setup:

python test_mcp_connection.py

Run the test suite:

pip install -e ".[dev]"
pytest tests/ -v