Home / MCP / Memory MCP Server
Provides memory primitives for AI agents, enabling persistent context across conversations.
Configuration
View docs{
"mcpServers": {
"memory": {
"command": "npx",
"args": [
"-y",
"@modelcontextprotocol/server-memory"
]
}
}
}You use MCP servers to give AI agents secure, contextual access to tools and data sources. This guide walks you through using MCP servers locally or with a client, starting from simple reference servers to connect and test basic interactions with an MCP client.
Connect an MCP server to your AI client to start exploring tool access and data queries. Start a local server, then configure your MCP client to point at that server so your AI can call the server’s tools in natural language workflows. Use memory-focused sessions to test persistent memory across conversations, then expand to additional servers as needed.
Install and run a couple of reference MCP servers to begin testing. Below are concrete steps you can follow to start a memory server and a Git server from your development machine.
# Start Memory MCP Server (TypeScript, via npx)
npx -y @modelcontextprotocol/server-memory# Start Git MCP Server (TypeScript, via npx) - complete command shown
npx -y @modelcontextprotocol/server-githubIf you plan to run multiple MCP servers locally, keep each server in its own terminal session. Use an MCP client to connect to each server by specifying the appropriate command and arguments in your client configuration. For example, load the memory server by pointing your client to the memory MCP tooling and verify that calls return expected results before adding more servers.
When crafting your MCP client configuration, reference the start commands above to create executable entries for each server. Include any environment variables your servers require and ensure transports (stdio or http) are set according to the server you run.