Skip to content
Agentmart
Submit a tool

Mneme Memory

Cross-session memory for AI coding agents via six tools; local-first SQLite store.

Connector by modusensus · GitHub account verified by the MCP Registry

IndexedMITLocal (stdio)★ 176
Write a review

Install

Works with Claude Code, Codex, VS Code, Cursor, Windsurf, Claude Desktop. Read the repository before you give any tool access to your data.

npx agentmarket install connector/modusensus-mneme-memory

Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.

Get the weekly diff

The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.

We use your email only for this newsletter. Privacy

About

Cross-session memory for AI coding agents via six tools; local-first SQLite store.

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this server can do and what it asks for, from its tool list and install config. Checked Oct 25, 2026.

Tools
4 read · 8 write · 1 delete
Secrets
None asked for
Install scripts
Not scanned yet
Runs as
Node.js package

Tools by what they do 13

Read from the source code; the running server may list different tools.

Read 4 · Look things up without changing anything.

  • memory_get
  • memory_list
  • memory_runtime
  • memory_search

Write 8 · Create, change, send or run things.

  • memory_archive
  • memory_forget
  • memory_organize
  • memory_register_document
  • memory_save
  • memory_update
  • mo-run
  • set_scope_state

Delete 1 · Delete, drop, revoke or wipe things.

  • memory_delete

Tools (13)

Read from the source code; the running server may differ. Descriptions are the server’s own words; hidden characters are shown as [U+…].

  • memory_archive

    Archive a memory (hide it from active lists, search, injection and dream consolidation) or restore it. Archived entries stay in storage and are recoverable: pass archived=false to restore, and use memory_list with include_archived=true to find archived entries. Only archive when the user asks, or an entry is clearly stale/no longer relevant — do not archive proactively mid-session to tidy up.

  • memory_delete

    Permanently delete a memory entry.

  • memory_forget

    Stop a memory from being auto-injected and from appearing in searches and lists without deleting it. The entry stays in storage; pass forgotten: false to restore it.

  • memory_get

    Fetch one memory entry by ID and return its full content as text. Use after memory_list to read a specific entry.

  • memory_list

    List memory entries by type, high-importance first, then newest, paginated. Set include_archived=true to also list archived (hidden) entries so they can be located and restored.

  • memory_organize

    Close out a stretch of work: compare candidate memories against the store first, then land only what you decide to keep. Two steps through one tool. mode=dryRun (default) takes `candidates` and answers one verdict per candidate — exact / near / new — with the matched row when there is one; it writes NO memories, only an audit receipt. mode=apply takes that report's `run_id` plus one decision per candidate (save | discard) and may archive a row the report flagged as superseded (action=archive with that matched id). apply is rejected without a real run_id, and a report can be applied only once: …

  • memory_register_document

    Register an agent-authored document (research report, design doc, long digest) into memory as a pointer row. Write-ownership split: the full text stays agent-owned on disk — the pipeline never reads or rewrites it; the DB stores only the summary + doc_path + evidence. Validates the file exists (absolute path, non-empty regular file), intersects evidence with real memory ids (all-fabricated evidence is rejected; unknown ids are dropped and the row is tagged evidence_degraded), and dedupes: re-registering the same path or title supersedes the old row (the old file is never touched; content_histo…

  • memory_runtime

    Inspect or provision the LOCAL inference runtime that dsh-mneme's local (offline) embedding and rerank need: the transformers + onnxruntime dependency closure, hundreds of MB unpacked. It is deliberately NOT a default dependency — mneme's core stays lightweight and local vectorization is opt-in — so this tool is the hook that makes the heavy capability available on demand. Use action=status first: it is read-only and reports whether the runtime is ready plus what provisioning would cost. action=provision tries, in order: (1) adopting an existing copy from this profile's node_modules (no networ…

  • memory_save

    Persist one memory entry for future sessions (user preferences, project state, decisions). Call this when the user states a durable preference, a project decision is made, or a lesson is learned. Merges into an existing entry of the same type when the title matches.

  • memory_search

    Search the cross-session memory store for past context (solved problems, user preferences, project decisions). IMPORTANT: relevant cross-session memories are already injected into your context every turn — only call this when the injected memory block does NOT contain what you need, or you need to look something specific up. Substring-matches title/content/tags, and augments results with semantic (vector) recall + optional rerank when an embeddings provider is configured. Returns matching entries with source and timestamps.

  • memory_update

    Modify an existing memory entry (title, content, type, tags, importance).

  • mo-run

    Description is built at runtime.

  • set_scope_state

    Press or release the reversible scope gate for one scope (issue #471): while paused, that scope's memories stay stored and unchanged but are held back from injection and dream consolidation. This is not deletion and not forgetting — release with state=active and everything comes back exactly as it was, which is why it is the right tool for temporarily silencing a whole work stream rather than editing memories one by one. dimension is agent or workspace; scope_value is the label itself, or * for that dimension's global gate. Both switches (scopeState.enabled/enforce) are off by default, so sett…

Signals

Stars
176
Forks
22
Open issues
18
Contributors
18
Last commit
Oct 24, 2026 (6 days ago)
Last release
Oct 21, 2026 (8 days ago)

From GitHub (fetched Oct 24, 2026).

Details

Registry name
io.github.modusensus/mneme-memory
Version
0.1.1
Auth
No auth
Packages
npm mneme-memory
License
MIT
First seen
Oct 24, 2026
Last checked
Oct 24, 2026
Cite this page

Mneme Memory MCP server. Agent Mart. https://agentmart.dev/connectors/modusensus-mneme-memory (last updated 2026-10-24)