Show HN: Lapse - a notes app. but also shared memory space for your agents (MCP)

Lapse appeared on Hacker News as a Show HN submission under the HN AI Agents track: a notes app that also functions as a shared memory space for agents over MCP. At the time of the listing it had 2 points and 1 comment, with discussion at https://news.ycombinator.com/item?id=49179101. The pitch is that you talk, and the note is saved in your space and tagged.

On the technical side, Lapse is built to connect wherever the user already works: Claude, ChatGPT, Codex, Claude Code, opencode, a local LLM via Ollama, or any MCP client. Capture is not limited to a desktop chat UI; the author reports using it through an iPhone shortcut so voice becomes a note without a separate write-up step. Tagging is part of the save path, so items land in a shared space rather than as unstructured chat residue.

For engineers and builders running agents across tools, the useful idea is one memory surface instead of per-client context. Notes and agent-relevant state can live in the same place whether the client is a hosted model, a coding agent, or a local Ollama stack, as long as the client speaks MCP. That reduces the usual copy-paste and “which chat had that decision” problem when workflows jump between products.

Market context is the Show HN niche of personal agent infrastructure: not a full platform, but a thin shared layer between human notes and machine clients. Competitors in this space tend to be either pure note apps with no agent protocol, or agent stacks with memory that stays inside one vendor. Lapse’s claim is the middle ground—notes plus MCP-reachable shared memory—aimed at people already mixing Claude, ChatGPT, Codex, Claude Code, opencode, and local models.

Practical takeaway: if you already use MCP clients or an iPhone shortcut-driven capture habit, Lapse is worth a trial as a tagged shared space rather than another siloed notebook. What to watch next on the HN thread is whether early users report clean multi-client consistency, how tagging holds up under agent write load, and whether the single-comment discussion grows into concrete integration feedback.