Problem
The current Supermemory MCP server () exposes three tools: memory (save/forget), recall (search), and listProjects. There's no way for an AI agent to enumerate all stored memories — only the top-K results of a recall query, or save/forget one at a time.
This matters for several real workflows:
- "Show me everything you remember about project X" → today requires guessing the right recall query and hoping nothing gets dropped from the top-K.
- Auditing what the agent has stored ("list all the preferences you have on file for me").
- Bulk cleanup of stale or duplicate memories before re-saving.
- Diffing memories across two container tags.
Proposed
Add a new listMemories tool to apps/mcp/src/server.ts, parallel to recall but returning paged memory records:
{
"name": "listMemories",
"description": "Enumerate stored memories. Use this to audit what is on file before a save/forget operation, or to power a 'list everything' view.",
"inputSchema": {
"containerTag": "string (optional)",
"limit": "integer (default 50, max 200)",
"cursor": "string (optional, opaque pagination token)",
"sort": "string, one of [created_desc, created_asc, updated_desc] (default created_desc)",
"filter": "string (optional substring filter against memory content)"
}
}Returns:
{
"memories": [{ "id": "...", "content": "...", "createdAt": "...", "updatedAt": "..." }],
"nextCursor": "..."
}Why it fits
- Single-file change in
apps/mcp/src/server.ts (one server.registerTool("listMemories", ...) block + a thin SupermemoryClient method in client.ts).
- Reuses the existing
containerTag scoping and includeProfile semantics already used by recall.
- No new dependencies; the underlying search API already supports cursor pagination — just needs a non-query entry point.
Use case
An agent doing a periodic "clean up my memory" task: list all memories, surface the ones the user might want to drop, then call memory with action: "forget" on the chosen IDs. Today this loop is impossible because there's no list step.
Happy to send a PR against main if the maintainers are open to it — would be ~80-150 LOC.
Problem
The current Supermemory MCP server () exposes three tools:
memory(save/forget),recall(search), andlistProjects. There's no way for an AI agent to enumerate all stored memories — only the top-K results of arecallquery, or save/forget one at a time.This matters for several real workflows:
Proposed
Add a new
listMemoriestool toapps/mcp/src/server.ts, parallel torecallbut returning paged memory records:{ "name": "listMemories", "description": "Enumerate stored memories. Use this to audit what is on file before a save/forget operation, or to power a 'list everything' view.", "inputSchema": { "containerTag": "string (optional)", "limit": "integer (default 50, max 200)", "cursor": "string (optional, opaque pagination token)", "sort": "string, one of [created_desc, created_asc, updated_desc] (default created_desc)", "filter": "string (optional substring filter against memory content)" } }Returns:
{ "memories": [{ "id": "...", "content": "...", "createdAt": "...", "updatedAt": "..." }], "nextCursor": "..." }Why it fits
apps/mcp/src/server.ts(oneserver.registerTool("listMemories", ...)block + a thinSupermemoryClientmethod inclient.ts).containerTagscoping andincludeProfilesemantics already used byrecall.Use case
An agent doing a periodic "clean up my memory" task: list all memories, surface the ones the user might want to drop, then call
memorywithaction: "forget"on the chosen IDs. Today this loop is impossible because there's no list step.Happy to send a PR against
mainif the maintainers are open to it — would be ~80-150 LOC.