dan-calin/shared-agent-memory

shared-agent-memory is a tool that gives all your AI coding agents a single shared memory so that what one agent learns, the others can immediately access. Without it, each new AI session starts from scratch and you have to manually copy context between agents by writing handoff files and pasting them around. This tool replaces that process with a searchable knowledge graph stored in a single JSON file on your machine. The technical approach is straightforward: it configures the memory MCP server, a standard memory system from the MCP protocol ecosystem, for each AI agent you use (Claude Code, Codex, or any other MCP-compatible tool) and points all of them at the same file. When an agent finishes a task, it saves a short note about what changed. When a new agent starts a task, it searches that memory for relevant prior context before doing anything. The memory is organized as named entities with short observations, plus relationships between them, so searching finds the right facts quickly without reading an ever-growing document. An optional coordination layer handles a second problem: two agents editing the same file at the same time and overwriting each other. You can turn it on with one command, after which agents can claim the files they intend to edit. The claims are advisory by default, meaning the system warns an agent if it is about to touch a file claimed by another agent but does not block the edit. A stricter mode that refuses the edit until the claim is released is also available. Claims expire automatically after two hours so a crashed session does not leave a file permanently locked. Installation uses one npx command that auto-detects which agents you have installed and configures the MCP server for each. Agents then use the memory automatically at the start and end of each task with no manual steps required. The full README is longer than what was shown.