MCP SERVERPersistent memory for an agent in one SQLite file — layered by what kind of knowledge it is, and versioned when a fact changes.
LABSCO SUMMARY
A journal entry from last Tuesday and a coding rule that never changes do not belong in the same bucket: the episodic layer decays on a 30-day half-life while procedural memories are loaded at startup and rarely move. Facts are superseded rather than overwritten, so the chain from “Team uses React 17” to “Team uses React 19” is still readable when you need to know what changed and when.
A local memory store for MCP clients. Memories go into four layers — episodic for events, semantic for facts, procedural for rules, resource for reference material — each with its own access pattern and decay, and everything lives in a single SQLite database on your machine.
memory_add with layer, entity, confidence, importance and an optional TTL, plus memory_search, memory_update, memory_delete and memory_inspectmemory_session_start returns an ID to attach memories to, memory_session_end takes a summary, memory_session_list shows what happened whenmemory_health — expired entries, orphaned superseding chains, stale and never-accessed memories, low-confidence entries, with optional clean-upmemory_export to JSON, Markdown or a compact claude-md format, filtered by layer, scope or date rangememory://stats, memory://recent, memory://layer/{layer}, memory://entity/{name} — and three prompts: recall, context-load, journalnpm install -g mnemon-mcp, then point your client at the mnemon-mcp command over stdio. No account, no key, no cloud: the database is created on first run at ~/.mnemon-mcp/memory.db, and MNEMON_DB_PATH moves it elsewhere. Vector and hybrid search are optional and bring-your-own-key — set MNEMON_EMBEDDING_PROVIDER to openai with MNEMON_EMBEDDING_API_KEY, or to ollama to keep it on your machine. With neither set, search stays on full text and needs no model and no download.
One command — npm install -g mnemon-mcp
USE CASES
Put it in the procedural layer, where it is loaded at startup rather than found by search.
Supersede the old fact instead of deleting it — search returns the new one and the chain stays inspectable.
Open a session, group what happens into it, then close it with a summary of what was accomplished.
Search by entity across every layer, narrowed by date or confidence.
Report expired entries, orphaned chains and never-accessed memories, then remove what no longer holds.
Counted once with only startup setup. Other versions, modes or optional credentials may expose a different list.
memory_addAdd a new memory to the persistent store. Supports 4 cognitive layers: episodic (events/sessions), semantic (facts/concepts), procedural (rules/workflows), resource (reference material). Automatically supersedes previous entries from the same source_file.
memory_deletePermanently delete a memory by ID. Cleans up superseding chain references: re-activates predecessor if one exists.
memory_exportExport memories to JSON, Markdown, or claude-md (compact LLM-optimized) format. Supports filtering by layer, scope, date range. Returns the exported content as a string.
memory_healthDiagnostic health report on the memory store. Returns expired entries, orphaned superseding chains, stale/never-accessed memories, low-confidence entries, and per-layer stats. Use cleanup=true to garbage-collect expired entries.
memory_inspectInspect memory details or layer statistics. Without id: returns aggregate stats per layer (total, active, superseded, avg_confidence, top_entities). With id: returns the full memory row and optionally its history chain.
memory_searchFull-text search across all memory layers using FTS5. Supports layer/entity/date/scope filtering. Returns scored results with snippets. Superseded entries excluded by default.
memory_session_endEnd an active session. Optionally attach a summary of what was accomplished. Returns duration and count of memories created during the session.
memory_session_listList recent sessions with their metadata. Supports filtering by client, project, and active-only. Returns memories count per session.
memory_session_startStart a new agent session. Returns session ID that can be passed to memory_add(session_id=...) to group memories. Use at the beginning of a work session.
memory_updateUpdate an existing memory. Use supersede=true to create a versioned replacement (preserves history chain). Use supersede=false (default) to update fields in place.
BEFORE YOU INSTALL
10 tools when we connected.
CONTEXT COST
Rated by the size of the 10 tool definitions that ride along on every turn, not by how many there are
INSTALLATION
npm install -g mnemon-mcp{
"mcpServers": {
"mnemon-mcp": {
"command": "mnemon-mcp"
}
}
}git clone https://github.com/nikitacometa/mnemon-memory-mcp.git
cd mnemon-memory-mcp && npm install && npm run buildUse the full path to the compiled entry point:
{
"mcpServers": {
"mnemon-mcp": {
"command": "node",
"args": [
"<YOUR_PATH_TO_MNEMON_MCP>/dist/index.js"
]
}
}
}MCP servers
TSchonleber
GuyMannDude
sgx-labs
Thezenmonster