ProjectContext - MCP Server

Improved Successor of AgentMemory

A high-performance MCP (Model Context Protocol) server providing long-term memory storage with semantic and keyword search capabilities, along with a structured agenda engine for task management.

Features

• Fast Semantic Search: Uses fastembed with BAAI/bge-small-en-v1.5 for fast startup and low memory usage
• Hybrid Search: Combines keyword (FTS5) and vector search using Reciprocal Rank Fusion (RRF)
• Agenda Engine: Task management with full-text search for plans and todo lists
• MCP Prompts: Specialized workflows for onboarding, feature planning, and memory maintenance
• Persistent Storage: SQLite-based storage with sqlite-vec extension
• Sub-200ms Queries: Keeps embedding model in memory for fast response times
• MCP Native: Exposes tools, resources, and prompts natively for AI agents

Installation

# Clone the repository
git clone <repo-url>
cd projectcontext

# Install dependencies with uv
uv sync

# Or install globally
uv pip install -e .

Usage

Running the Server

# Run directly
projectcontext

# Or with uv
uv run projectcontext

MCP Configuration

Add to your MCP client configuration (e.g., mcp.json):

{
  "mcpServers": {
    "projectcontext": {
      "command": "uv",
      "args": ["run", "projectcontext"],
      "cwd": "/path/to/projectcontext"
    }
  }
}

Or using the installed script:

{
  "mcpServers": {
    "projectcontext": {
      "command": "projectcontext"
    }
  }
}

MCP Tools

Memory Engine Tools

• save_memory: Save a memory with category, topic, and content.
• query_memory: Search memories using hybrid semantic/keyword search.
• update_memory: Modify an existing memory by ID.
• delete_memory: Remove a memory by ID.

Agenda Engine Tools

• create_agenda: Create a new multi-step plan or todo list.
• list_agendas: Show all active or inactive agendas.
• get_agenda: Retrieve detailed task information for a specific agenda.
• search_agendas: Search plans by title or description.
• update_task: Mark tasks as completed or pending.
• update_agenda: Modify agenda metadata or add new tasks.
• delete_agenda: Remove inactive agendas.

MCP Resources

projectcontext://usage-guidelines

Provides comprehensive documentation for AI agents on how to effectively use the Memory and Agenda engines, including categorization best practices and hallucination prevention.

projectcontext://schemas/{tool}

Provides the JSON schema for a specific tool. This is useful for AI agents to understand the required and optional parameters for each tool.

MCP Prompts

ProjectContext includes built-in prompts to guide AI agents through complex workflows:

• setup_project_context: Templates for initializing a new project's tech stack, goals, and conventions.
• plan_feature_implementation: A structured workflow for searching existing context and creating a multi-step agenda for new features.
• summarize_and_remember: Distills conversation history into structured memories while avoiding duplicates.
• debug_with_history: A troubleshooting workflow that leverages past bug_fix memories and system context.
• maintain_memory_health: A proactive maintenance workflow for identifying and cleaning up outdated or redundant information.

Architecture

Technology Stack

• Framework: FastMCP (Python MCP library)
• Embeddings: fastembed (BAAI/bge-small-en-v1.5, 384-dim)
• Database: SQLite with sqlite-vec and FTS5 extensions
• Communication: JSON-RPC over stdio

Storage Location

The databases are stored in the .ctxhub/ directory in the git root (or current working directory).

• memory.sqlite: Memory Engine database
• agenda.sqlite: Agenda Engine database

Development

Project Structure

projectcontext/
├── src/
│   └── projectcontext/
│       ├── __init__.py      # Package initialization
│       ├── server.py        # MCP Server (Tools, Prompts, Resources)
│       ├── memory.py        # Memory Engine Logic
│       ├── agenda.py        # Agenda Engine Logic
│       └── database.py      # Database Utilities
├── pyproject.toml
└── .ctxhub/                 # Databases

Testing

# Quick start: runs main tests and offers to start server
./quickstart.sh

# Run specific tests manually
uv run python tests/test_server.py
uv run python tests/test_updates.py

MCP Inspector

npx @modelcontextprotocol/inspector uv run projectcontext

License

GPL-3.0-or-later