projectmemory

ProjectMemory

ProjectMemory is an MCP (Model Context Protocol) server that provides persistent storage for conversation context information using SQLite. This allows LLMs to remember and retrieve relevant information from past interactions.

Overview

ProjectMemory implements two MCP tools:

1. save_context - Saves context (conversation snippets, inputs, outputs) to a persistent store
2. retrieve_context - Retrieves relevant context based on semantic search

The service handles:

• Summarizing text to extract key information
• Creating embeddings for semantic search
• Storing context with metadata in SQLite
• Searching for relevant context using vector similarity

Components

• Vector Package: Utilities for embedding operations and vector similarity
• Summarizer Package: Text summarization capabilities
• ContextStore Package: SQLite-based persistent storage
• Server Package: MCP server implementation
• Logger Package: Structured logging system

Getting Started

Prerequisites

• Go 1.20+
• SQLite

Configuration

Create a .projectmemoryconfig file in the project root:

{
  "models": {
    "provider": "mock",
    "modelId": "mock-model",
    "maxTokens": 1024,
    "temperature": 0.7
  },
  "database": {
    "path": ".projectmemory.db"
  },
  "logging": {
    "level": "INFO",
    "format": "TEXT"
  }
}

Running the Server

From the project root:

go run cmd/project-memory/main.go

Using as a Library

ProjectMemory can be used as a library in your Go applications in multiple ways:

1. Direct Component Usage - Directly use the core components for maximum control
2. Helper Functions - Use the CreateComponents helper for easier initialization
3. High-Level API - Use the Server API for simplified operations

These approaches allow you to integrate ProjectMemory with your existing MCP server without conflicts. For detailed instructions and examples, see our Library Usage Guide and our comprehensive Embedding Guide.

Quick Example

import (
    "github.com/localrivet/projectmemory"
    "github.com/localrivet/projectmemory/internal/contextstore"
    "github.com/localrivet/projectmemory/internal/summarizer"
    "github.com/localrivet/projectmemory/internal/vector"
)

// Initialize components
store := contextstore.NewSQLiteContextStore()
store.Initialize(".projectmemory.db")
defer store.Close()

summ := summarizer.NewBasicSummarizer(summarizer.DefaultMaxSummaryLength)
summ.Initialize()

emb := vector.NewMockEmbedder(vector.DefaultEmbeddingDimensions)
emb.Initialize()

// Now you can use these components directly in your code
// For example, to store context:
testText := "This is a test context to save."
summary, _ := summ.Summarize(testText)
embedding, _ := emb.CreateEmbedding(summary)
embeddingBytes, _ := vector.Float32SliceToBytes(embedding)
id := projectmemory.GenerateHash(summary, time.Now().UnixNano())
store.Store(id, summary, embeddingBytes, time.Now())

For a complete example of integrating with an existing MCP server, see examples/embed-in-mcp/main.go.

API Reference

Tool: save_context

Request:

{
  "context_text": "The text to save in the context store"
}

Response:

{
  "status": "success",
  "id": "generated-unique-id"
}

Tool: retrieve_context

Request:

{
  "query": "The query to search for",
  "limit": 5
}

Response:

{
  "status": "success",
  "results": ["Matching context entry 1", "Matching context entry 2", "..."]
}

Tool: delete_context

Request:

{
  "id": "context-entry-id-to-delete"
}

Response:

{
  "status": "success"
}

Tool: clear_all_context

Request:

{
  "confirmation": "confirm"
}

Response:

{
  "status": "success"
}

Tool: replace_context

Request:

{
  "id": "context-entry-id-to-replace",
  "context_text": "The new text to replace the existing context"
}

Response:

{
  "status": "success"
}

Documentation

• Installation Guide
• Configuration Reference
• API Reference
• Development Guide
• Architecture Documentation
• Library Usage Guide
• Embedding Guide - Comprehensive guide for embedding in your application

License

MIT License