MemStore

Teams that use AI agents daily on tasks like coding lose context between sessions, and one person's agent has no way to know what a teammate's agent already decided, tried or committed to. The sample rollup included on the page shows the cost concretely: a PR blocked because scoring criteria were never given, a beta user stuck at 'accepted' because login instructions were never sent, overdue details owed to an investor, and several workstreams (ontology pull requests, EC2 cost review, a Fathom integration) left unowned for weeks. Slack decisions also need to be manually recorded, since nothing captures them automatically. Without a shared record, teams rely on standups and memory to catch blocked work and overdue commitments, which the page shows failing repeatedly.

MemStore is an MCP server exposing three tools agents call as they work: checkin (returns where work stands and what teammates are doing), record (logs decisions, progress and commitments to the user's store), and recall (pulls answers from the user's or teammates' stores, with the source attached). It works with Claude, Claude Code, ChatGPT, Codex and other MCP clients, and connects to GitHub and Fathom plus a Claude chat importer for past conversations. Each morning it sends a personalized rollup email summarizing what changed, what's blocked, and who is waiting on whom, grounded in a stated count of memories from the user's own store and teammates' stores. Teammates' agents can search a user's store, including personal material recorded there, since the product is built for teams that share context by default.