Repository navigation
This release adds node renaming, makes it safe for several MCP clients to share one memory file, and fixes the SchemaManager and MemoryViewer tools.
Upgrading
git pull
npm installThen restart your MCP client. Tool names, schema files and the memory file format are unchanged.
Node.js 22 or newer is now required. Node 18 and 20 have reached end-of-life, and the tests run on 22 and 24.
Server
- Nodes can be renamed.
update_nodesacceptsnewName. The node's edges follow, and so does metadata that refers to it by name: its ownname: ...entry, and entries such ascurrentLocation: Townon nodes linked to it. - Two MCP clients can safely share one memory file. Each change takes a short-lived lock file (
memory.json.lock) next to the memory file, so two MemoryMesh processes can no longer overwrite each other's changes. - Names are trimmed when a node is created, so
" Mara "can no longer be added next to"Mara". - Tools carry MCP annotations saying whether they only read, only add, or change existing data. Clients use these to decide when to ask for confirmation.
- Tool results use compact JSON instead of indented JSON, so every result takes less of the AI's context.
read_graphon a large graph benefits most. The format is otherwise the same: a one-line summary, a blank line, then the data.
SchemaManager and MemoryViewer
- SchemaManager checks a schema the way the server does before generating or saving it, so it no longer produces schemas that MemoryMesh skips. A description is required for the schema and every property, a relationship needs an edge type, an enum needs at least one value, and the name must be usable as a tool name.
- SchemaManager no longer raises an error on every keystroke in a property name field. The hint it was trying to show now works, and flags duplicate names and
metadata. - MemoryViewer skips lines it cannot read and says which, instead of refusing to show the whole file because of one bad line.
Full details are in the changelog.