Give MCP agents local shared memory with MARM
Use MARM Memory to give Claude Code, Cursor, Codex, Gemini, and other MCP clients a local-first shared memory server with sessions, semantic recall, notebooks, summaries, and code graph context.
npx skills add agentskillexchange/skills --skill give-mcp-agents-local-shared-memory-with-marm
Use MARM Memory when an operator needs coding agents or MCP clients to remember project decisions, session history, reusable facts, and codebase structure across runs. The agent/operator installs the MARM MCP server, starts it over HTTP or stdio, connects one or more clients, and uses its memory and code graph tools to write session logs, retrieve prior context, index repositories, inspect symbols, and summarize long-running work. Invoke it instead of relying on a single chat transcript, ad hoc Markdown notes, or separate memory and code-index tools when several agents need local, private, shared recall. The scope boundary is persistent MCP-accessible project memory and code context for agent work; it is not a general vector database, full agent runtime, or hosted personalization product.
What this skill actually does
Inputs and prerequisites: MARM MCP server, Python or Docker runtime, SQLite-backed local storage, MCP-compatible clients such as Claude Code, Cursor, Codex, Gemini, Qwen, or VS Code agents.
Setup notes: Install the MARM MCP server from the upstream package or Docker image, configure it for stdio or HTTP MCP transport, point each MCP client at the server, then use the bundled memory, session, notebook, summary, and code graph tools during project work.
Source and verification boundary: use https://github.com/Lyellr88/marm-memory as the canonical reference before running the workflow; keep commands, API calls, CLI usage, and generated outputs reviewable against that upstream source.
Framework fit: publish this as a MCP workflow only when the operator can invoke the documented toolchain directly, rather than treating the upstream project as a generic product listing.