SkillFoxx

Mneme MCP

Long-term memory for DeepSeek Harness with an MCP server: the same memories reachable from Claude Code, Cursor and Codex

  • 223 ★ on the repository
  • Last commit
  • MIT
MCP server

Extended access

Why this level

  • The server reads and changes your long-term memory, and the MNEME_TOKEN opens it to any client
  • Memory holds preferences and project decisions that may be sensitive
All reasons and checks

slow-stack/mnemeAuthor:slow-stack

Install

In your terminal, with SkillFoxx CLI

npx skillfoxx add mcp/mneme

Detects the agents on your machine, shows the access level and pins the version.

Other ways to install

Access and checks

How we review

Extended accessThe tool runs code or makes network calls and only reads external data.

  • The server reads and changes your long-term memory, and the MNEME_TOKEN opens it to any client
  • Memory holds preferences and project decisions that may be sensitive

This is third-party code. Review the repository files before installing.

What it does

dsh-mneme is a cross-session memory plugin for DeepSeek Harness (DSH). Memory is stored on your machine in SQLite with Markdown mirrors: projects, preferences and decisions survive closing the window. The background autoDream process deduplicates and merges entries, archives rarely used ones and freezes contradicting memories until you decide. Recall has four channels: keyword, vector, BM25 and entity. For other clients the plugin includes a dependency-free stdio MCP server, the `mneme-memory` package with the `mneme-mcp` command. It exposes six tools: `memory_save`, `memory_search`, `memory_list`, `memory_get`, `memory_update` and `memory_delete`. Data goes through the plugin's local API at `127.0.0.1:8790` and the token is created in the DSH panel under External API.

Who it is for. DeepSeek Harness users who need shared memory across sessions and people who want to attach the same memory to Claude Code, Cursor or Codex.

Good fit when

  • Two days later the agent recalls that you picked Astro for the blog rework and does not ask again
  • Two memories contradict each other and you settle the conflict in a queue, comparing them side by side
  • Memory is split by agent and workspace so projects do not mix
  • The same memories are needed in Claude Code through the `mneme-mcp` command

Not a fit when

  • You do not run DeepSeek Harness: the MCP server goes through the plugin that runs inside DSH
  • You need cloud sync across devices: memory stays on the local machine
  • You need a stable version: the README marks pre-release builds such as 0.14.0-beta

Example request

Search memory for what we decided about the blog rework, show the entry and save the clarification that Astro was chosen for it

MCP

Transport
stdio
Authentication
API key

Environment variables

MNEME_TOKEN required, secret
Token from the DSH panel under External API for access to the plugin's local API

Limitations

The MCP server works only while DeepSeek Harness runs with the plugin installed, and data goes through the local API at `127.0.0.1:8790`. The token has to be created in the DSH panel under External API and passed in `MNEME_TOKEN`. Installing the plugin needs Node.js 22 or newer. Recall uses an external embeddings provider (`openai`) by default and works fully offline after `embedProvider` is set to `local`. The version at the time of checking is a pre-release. Documentation is in Chinese and English and the license is MIT. The old names `dsh-mneme-mcp` and `DSH_MNEME_TOKEN` are still supported.

How to disable. Remove the server from the client and revoke the token in DSH. The plugin is removed with `dsh plugin --profile web remove @modusensus/dsh-mneme`, and the memory file in `~/.dsh/memory` stays and is deleted by hand.

FAQ

How do I connect Mneme to Claude Code?

Add a `mneme-memory` server with the `mneme-mcp` command and the `MNEME_TOKEN` environment variable to the project's `.mcp.json`. Take the token from DSH under External API.

Which tools does the MCP server expose?

Six: `memory_save`, `memory_search`, `memory_list`, `memory_get`, `memory_update` and `memory_delete`.

What if the package is not installed globally?

Run through npx: the `npx` command with the arguments `-p mneme-memory mneme-mcp`.

Can it work offline?

Yes, after `embedProvider` is set to `local` recall uses local embeddings and needs no API key. Without embeddings keyword and BM25 search still work.

Where is the data stored?

In a SQLite file at `memoryDir`, by default `~/.dsh/memory`, with Markdown mirrors. Removing the plugin does not touch it.

Discussion

No questions or reviews yet.

MCP servers

Agent memory as a graph of linked facts: decisions and errors outlive the session, with no vector database or paid API calls

MCP serverExtended access244

A long-term memory server for agents with a Mem0-compatible API, hybrid search, its own MCP server and Docker deployment.

MCP serverExtended access114

A self-hosted MCP memory server that keeps preferences and decisions across conversations in Claude Code, Cursor and other clients

MCP serverExtended access283
Official

Long-term memory for Claude Code, Cursor and Codex: the agent saves decisions and finds them again in later sessions

MCP serverExtended access66

MemMesh, MCP-сервер памяти в одном бинарнике на Rust, хранит заметки агента на машине и ищет по смыслу без внешней базы.

MCP serverExtended access358

README in short

The Chinese and English text promises memory that tidies itself: it covers scenarios, trust in background processes, a five-minute setup, settings, upgrades and multi-agent support. A separate section describes MCP for other clients.

README badge

Report a problem