NMEMORY MCP Server

Hermetic local memory for coding agents: provenance-mandatory capture, honest recall, zero sockets.

Local serverstdioPython

What is the NMEMORY MCP server?

Connect NMEMORY to Claude, Cursor or any other MCP client and it stops being a tab you switch to. Hermetic local memory for coding agents: provenance-mandatory capture, honest recall, zero sockets. The nmemory mcp server is what makes that connection.

What the server does

Your coding agent forgets you between sessions, so you spend every morning re-explaining decisions it was there for. Bolt a memory onto it and you usually trade that for something worse: asked a question it has nothing solid for, the memory hands back something plausible anyway, and your agent goes and edits your code with it.

Installation

Setup follows the usual MCP pattern — install or clone the server, register it in your client's configuration file, restart the client.

Available tools

The toolset is worth reading before you wire it up, because it tells you what the integration is really for:

  • memory_ingest — capture with a birth certificate: no source + anchor, no storage
  • memory_retrieve — recall as evidence: grounded, missing_evidence, or an honest abstain
  • memory_digest — session-start projection: what you know, what's ready, what's blocked
  • memory_relate — the five edges you reach for daily, out of nine declared: supersedes, derived_from, witnesses, blocks, and falsifies (a disproven fact stops
  • memory_forget — tombstones with audit, never silent deletion
  • memory_extract — text → candidate memories over the closed 10-kind set; advisory, stores nothing
  • memory_classify — kind / scope / authority / taint labels; optionally persisted as a sidecar
  • memory_import — one-shot import of native sources (CLAUDE.md, AGENTS.md, memory dirs); born tainted
  • memory_get — one full capsule by id, with GET-only fact time, relations,
  • memory_list — compact index with project fences
  • memory_bootstrap — cold-start pack: your constraints FIRST (never capped), the one next action, decisions, traps — in ≤1500 tokens
  • memory_alias — teach recall synonyms the store then honors

Worth knowing first

  • It runs with your machine's permissions. That is convenient and also the reason to think about what you point it at before you approve a tool call.
  • With 14 tools registered it takes up a noticeable share of the context window; turn it off in projects that never touch NMEMORY.
  • Keep per-call confirmation enabled while you learn its behaviour; it is the cheapest safeguard you have.

Where it fits

Plenty of knowledge and memory servers cover similar ground. The differences that matter in practice are scope of access and how much setup stands between you and a working tool call. NMEMORY's toolset — memory_ingest, memory_retrieve, memory_digest and 11 more — is a fair guide to whether it matches your workflow. It is maintained by menot-you; worth a glance at recent repository activity before you build anything load-bearing on it.

We check each listing at SyncDev against the project's documentation before it goes live — if something here drifts out of date, it is a bug worth reporting.

Available tools

ToolWhat it does
memory_ingestcapture with a birth certificate: no source + anchor, no storage.
memory_retrieverecall as evidence: grounded, missing_evidence, or an honest abstain.
memory_digestsession-start projection: what you know, what's ready, what's blocked.
memory_relatethe five edges you reach for daily, out of nine declared: supersedes, derived_from, witnesses, blocks, and falsifies (a disproven fact stops grounding recall, but the evidence stays). The other four — proposes, part_of,
memory_forgettombstones with audit, never silent deletion.
memory_extracttext → candidate memories over the closed 10-kind set; advisory, stores nothing
memory_classifykind / scope / authority / taint labels; optionally persisted as a sidecar
memory_importone-shot import of native sources (CLAUDE.md, AGENTS.md, memory dirs); born tainted
memory_getone full capsule by id, with GET-only fact time, relations,
memory_listcompact index with project fences
memory_bootstrapcold-start pack: your constraints FIRST (never capped), the one next action, decisions, traps — in ≤1500 tokens
memory_aliasteach recall synonyms the store then honors
memory_vectorattach caller-fed embeddings (optional cosine lane; no embedder inside)
memory_visualdeterministic Mermaid projections (dag / relations / tiers / sessions), plus an MCP Apps view

Example prompts to try

  • Use NMEMORY to memory ingest.
  • Use NMEMORY to memory retrieve.
  • Use NMEMORY to memory digest.

Frequently asked questions

It connects NMEMORY to MCP-compatible AI assistants such as Claude and Cursor, exposing 14 tools (memory_ingest, memory_retrieve, memory_digest, and more) that the assistant can call on your behalf. Instead of copying data back and forth by hand, the assistant works with NMEMORY directly.