MCP SERVER · OFFICIAL PROJECTOne memory layer several agents share — save and recall by meaning, keep per-project state, TODOs and a timeline, and run a small expense ledger alongside it.
LABSCO SUMMARY
Saving and recalling is the easy part; the rest is what makes it usable inside an agent. Recalls carry explicit item and token budgets, so a memory layer cannot quietly eat the context window. get_project_context returns state, TODOs, decisions and timeline as one bounded pack, which is what resuming a session actually needs. Deletions record a reason and can be undone, and explain_memory answers the question most memory tools cannot: why did this come back for that query.
A hosted memory service reached over MCP. It stores memories that can be recalled by meaning later, keeps a scoped working state and timeline per project, tracks TODOs, and includes a ledger for expenses — all against one account, so the memories follow you between clients.
A hosted service and an XMemo account. Connect over Streamable HTTP at https://xmemo.dev/mcp with the bearer token in an Authorization header, or run the local stdio proxy from @xmemo/client on Node.js 20 or newer with XMEMO_KEY set — the key is optional for discovery and required to execute anything. Optional headers attribute activity to an agent family, a local installation and a named device, and XMEMO_URL points at a different base URL when you are not using the default.
One command plus a key — npm install -g @xmemo/client, then supply credentials
USE CASES
One call returns the working state, open TODOs, recorded decisions and recent timeline, capped so it fits the context you have.
Write the current task, the next action and any blocker into a scoped state record, and log the decision that got you there.
recall for a light lookup, recall_context when several memories are needed and the item and token limits have to be explicit.
Create the action item with a due time, list what is still open, and close it when it is done.
explain_memory gives the reason a memory exists or matched, and the version list shows how it got to its current form.
forget records the reason it went, and restore_memory brings it back when the reason turns out to be wrong.
Counted once with only startup setup. Other versions, modes or optional credentials may expose a different list.
add_expenseCreate one XMemo Ledger transaction and backing memory. Requires memory:write; it records a new transaction or reuses a semantic duplicate, and never deletes Ledger records. Use it for a purchase, income, refund, or transfer; use list_ledger_transactions or get_monthly_ledger_summary for reads. amount must be positive; transaction_type defaults to expense; blank transaction_date uses today's UTC date.
complete_memory_todoMark a TODO/action item completed.
create_memory_todoCreate a TODO/action item with an optional due time.
explain_memoryExplain why a memory exists or matched a query.
forgetPermanently delete a memory by target.
get_mcp_identityCheck XMemo connection status and the connected account/agent.
get_monthly_ledger_summarySummarize Ledger totals by month and currency.
get_project_contextRead one authorized project's bounded context pack: state, TODOs, decisions, timeline, recent memories, and optional durable recall. Requires memory:read; it does not mutate project memories, and access is audit-logged. Use an exact project_id for a whole-project snapshot; otherwise use recall_context. max_items/max_tokens bound the whole pack; recent_hours affects only timeline; durable_query requires include_durable_context.
get_timelineRead authorized timeline events newest first. Requires memory:read and makes no memory changes. Use it for recent history or session resumption; use recall_context for semantic multi-memory context. limit is clamped to 1-500; session_id and event_type are exact filters.
list_ledger_transactionsShow XMemo Ledger records.
list_memory_todosList open or completed TODO/action items.
list_memory_versionsList available versions for a memory.
memory_statsShow aggregate statistics for XMemo memories.
recallRecall the most relevant saved memories before answering.
recall_contextRead a multi-memory context pack. Requires memory:read and does not change content. Use when many memories need explicit budgets; use recall for lightweight answer or get_project_context for project snapshot. max_items/max_tokens bound rendered output.
record_eventRecord a significant session event, milestone, or decision.
rememberSave a memory so it can be recalled in future conversations.
restore_memoryRestore a previously deleted memory.
update_memoryUpdate the content or metadata of an existing memory.
update_stateCreate or replace one scoped working-state record for resuming a task, next action, or blocker. Requires memory:write; it versions that state slot and refreshes its expiry without deleting other memories. Use remember for durable facts or record_event for history. Provide content or a structured state field; ttl_seconds=0 means no expiry.
BEFORE YOU INSTALL
20 tools when we connected.
CONTEXT COST
Rated by the size of the 20 tool definitions that ride along on every turn, not by how many there are
INSTALLATION
npm install -g @xmemo/clienthttps://xmemo.dev/mcpAdd this hosted endpoint to your MCP client using the credentials listed under Requirements.
xmemo setup codexxmemo setup cursorxmemo setup copilotxmemo setup geminixmemo setup antigravityxmemo setup openclawxmemo setup hermesxmemo setup kiroxmemo setup grok{
"mcpServers": {
"XMemo": {
"type": "streamable-http",
"url": "https://xmemo.dev/mcp",
"headers": {
"Authorization": "Bearer ${XMEMO_KEY}"
}
}
}
}Values written as ${…} are filled in only by clients that expand variables.
Clients spell the "type" value differently: Claude Code and VS Code use "http", Cline uses "streamableHttp", and Roo Code uses "streamable-http". Claude Desktop does not accept a URL entry in its config file; remote servers go under Settings → Connectors.
{
"mcpServers": {
"XMemo": {
"command": "npx",
"args": [
"-y",
"--package",
"@xmemo/client@latest",
"xmemo-mcp"
]
}
}
}This block lists the tools without a credential, but running any of them needs one: sign in once with xmemo login (after npm install -g @xmemo/client), or expose XMEMO_KEY through your platform's secret manager.
MCP servers
