Compose the context your AI reads from.
Mnema turns your docs, instructions, and snippets into structured flows. Claude pulls each step in order, with your instructions between them — so your AI reasons through your context the way you'd brief a teammate.
Free forever for personal use · No credit card
How Mnema syncs
Every edit is captured by Yjs and broadcast through Hocuspocus to all connected clients within 50ms. Persistence flushes to Postgres on a 3-second debounce, batching rapid edits without losing intermediate state.
MCP read path
Claude calls
get_doc
on the MCP server, which returns the markdown of the doc as it exists
right now.
Flows turn docs into sequences.
- 01Architecture overviewRead this first to understand the system shape.doc
- 02Code conventionsReference these conventions for any code change.doc
- 03Active ticketsAddress these issues before suggesting new work.doc
Sequence content.
Drag docs from your workspace into a step list. Every step is one chunk of context the agent reads.
Annotate each step.
Add an instruction in your own words. Claude reads the instruction before reading the doc.
Claude pulls each step in order.
No giant context dump. Each step is a separate MCP call, so the agent reasons between them.
Compose, don't dump.
A flow is a sequence of content steps with your instructions. Claude reads each step in order — context delivered the way a teammate would brief one.
Every doc, addressable.
Your workspace is a content library. Docs, snippets, instructions — each becomes a step in any flow. One source, many compositions.
Native to MCP.
Flows expose as MCP tools. Claude (or any MCP-capable agent) calls them directly. No uploads, no copy-paste, no stale context.
Live editing, live context.
Edit a doc and every flow that references it updates instantly. The context Claude reads at 3:14 is the doc your teammate updated at 3:13.