Context Memory MCP Server

A Model Context Protocol (MCP) server that turns your local machine into a persistent "memory wallet" for AI copilots. It exposes tools for adding, listing, retrieving, updating, and deleting durable context memories, and is ready to plug into VS Code Insiders (or any MCP-compatible client).

Features

• Persistent storage backed by SQLite with WAL journaling for reliability and speed
• Rich metadata per entry: title, free-form content, importance (0-10), tags, and arbitrary JSON metadata
• Querying & filters for search, tag filtering, importance thresholds, and temporal windows
• Full lifecycle tools (memory.add, memory.list, memory.get, memory.update, memory.delete)
• Structured JSON outputs alongside human readable summaries for easy UI rendering
• Configurable data path via environment variables so you can keep memories anywhere (including synced drives)

Installation

cd context-memory-mcp-server
npm install
npm run build

During development you can use npm run dev to run the TypeScript source directly (note that the process will remain attached to the terminal waiting for MCP messages).

Running the server

The build step creates dist/index.js. You can launch the server with:

node dist/index.js

Smithery

If you prefer to orchestrate MCP servers with Smithery, this package now exposes both a CLI and programmatic entry points that Smithery understands:

• CLI: the npm bin context-memory-mcp-server starts the stdio transport, so Smithery can invoke it directly.
• Module export: importing startContextMemoryServer (or createContextMemoryServer) from context-memory-mcp-server returns the configured McpServer, which Smithery can wire up to any transport it launches.

Environment variables:

Variable	Description	Default
MCP_CONTEXT_MEMORY_HOME	Directory where the SQLite file should live	<cwd>/data
MCP_CONTEXT_MEMORY_DB	Full path to the SQLite file. Overrides MCP_CONTEXT_MEMORY_HOME.	<MCP_CONTEXT_MEMORY_HOME>/context-memory-wallet.db

When first launched the database (and intermediate directories) are created automatically.

VS Code Insiders integration

1. Install a client that speaks MCP inside VS Code Insiders. The official "AI Agents" preview (Settings → Features → AI → Enable AI Agent tools) or extensions like Claude for VS Code support custom MCP servers.
2. Add a server definition to your VS Code settings. Below is a sample for the Insiders AI Agent preview (uses the experimental mcp.servers setting):

// .vscode/settings.json or global settings
{
  "mcp.servers": {
    "context-memory": {
      "command": "node",
      "transport": "stdio",
      "args": [
        "/absolute/path/to/context-memory-mcp-server/dist/index.js"
      ],
      "env": {
        "MCP_CONTEXT_MEMORY_HOME": "/absolute/path/to/memory-data"
      }
    }
  }
}

For the Claude extension, use the analogous claude-dev.modelContextServers setting:

{
  "claudeDev.modelContextServers": [
    {
      "name": "context-memory",
      "command": "node",
      "args": [
        "/absolute/path/to/context-memory-mcp-server/dist/index.js"
      ],
      "env": {
        "MCP_CONTEXT_MEMORY_HOME": "/absolute/path/to/memory-data"
      }
    }
  ]
}

3. Reload VS Code. The client should detect the server via stdio and register the five memory tools. Each tool returns both a readable summary and a JSON block you can parse or render in the side panel.

Tool reference

Tool	Purpose	Required params	Optional params
memory.add	Persist a new memory entry	content	title, importance, tags[], metadata
memory.list	List & filter memories	–	search, tags[], minImportance, maxImportance, before, after, limit, offset
memory.get	Fetch a single memory by ID	id	–
memory.update	Update an existing entry	id	title, content, importance, tags[], metadata (any subset)
memory.delete	Delete an entry	id	–

All responses contain a Markdown-formatted summary plus a JSON payload (wrapped in triple backticks) to keep clients machine-friendly.

Development notes

• The project uses TypeScript with NodeNext module resolution; Node 18+ is required.
• Storage logic lives in src/storage.ts; schemas in src/schemas.ts; the MCP bootstrap is in src/index.ts.
• npm run build runs tsc and emits ESM output to dist/.
• You can adjust the DB location or swap backend storage by editing MemoryStore.

Roadmap ideas

• Support for embedding vectors / similarity scoring
• Optional encryption at rest for private workspace setups
• MCP resource templates so memories appear as browsable URIs inside compatible clients

Feel free to extend and adapt the server for your own copilots!