live-memory — Claude Code MCP server: codebase memory
Keeps what your agent learned about the repo between sessions. It asks memory instead of reading the same files again.
claude mcp add live-memory -- uvx shofer-live-memoryWhat live-memory does
live-memory runs as its own MCP server that builds up knowledge of your repository and outlives a single session. Your agent sends broad questions through one tool, ask_live_memory — where something lives, how a piece works, what calls it — and gets an answer it can act on right away. The server is read-only and confined to the workspace path: it cannot edit, create or execute anything.
Learning happens on its own. PostToolUse and FileChanged hooks take the file content your agent already reads or edits and hand it to the server, so nothing extra is scanned. Memory sits in a local JSON snapshot and tracks modifications and deletions. A single server covers every session in that workspace.
Start the server before Claude connects. The plugin only attaches to it and never launches it, so a stopped process means a connection error. On a Claude subscription it runs with no API key, and the memory model can be pointed at a local one or at any OpenAI-compatible endpoint.
What it helps with
- locate a module somewhere in a large repo
- understand an unfamiliar part of the code
- trace which code calls a function
- carry codebase understanding across sessions
- cut the premium model's spend on re-reading files
Example prompts for the agent
- Ask live-memory where the payment handler is defined
- What calls
buildIndex, and from where? - Show me what live-memory already knows about this repo
How to install live-memory
Platform CLI
claude mcp add live-memory -- uvx shofer-live-memoryuvx runs the command named after the package; if the README names another one, use uvx --from shofer-live-memory <command>
Platform CLI
codex mcp add live-memory -- uvx shofer-live-memoryuvx runs the command named after the package; if the README names another one, use uvx --from shofer-live-memory <command>
Config file
{
"mcpServers": {
"live-memory": {
"command": "uvx",
"args": [
"shofer-live-memory"
]
}
}
}uvx runs the command named after the package; if the README names another one, use uvx --from shofer-live-memory <command>
Config file
{
"servers": {
"live-memory": {
"type": "stdio",
"command": "uvx",
"args": [
"shofer-live-memory"
]
}
}
}uvx runs the command named after the package; if the README names another one, use uvx --from shofer-live-memory <command>
Platform CLI
gemini mcp add live-memory uvx shofer-live-memoryuvx runs the command named after the package; if the README names another one, use uvx --from shofer-live-memory <command>
Step by step, with commands for each platform: How to add an MCP server to Claude Code, Codex, Cursor, VS Code, Gemini CLI and more
Questions about live-memory
How do I install live-memory in Claude Code?
Start the server first: git clone https://github.com/shofer-dev/claude-code-live-memory, then cd claude-code-live-memory/deploy && ./install-service.sh. Inside a Claude session run /plugin marketplace add shofer-dev/claude-code-live-memory and then /plugin install live-memory@shofer-live-memory.
What does live-memory do?
It accumulates knowledge about your repository in a separate MCP server and answers the agent through the ask_live_memory tool. The agent gets an answer instead of re-reading files.
Does live-memory need an API key?
On a Claude subscription it runs without one. The memory model can also be switched to a local model or an OpenAI-compatible endpoint.
