SkillFoxx

Aura Memory MCP

Aura Memory gives an agent local long-term memory with trust levels, so the user own words outweigh text scraped from the web.

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

Extended access

Why this level

  • The Claude Code auto-capture hook stores the conversation verbatim, including any secrets that slip into it.
All reasons and checks

teolex2020/aura-memoryAuthor:teolex2020

Install

Manual install

python -m pip install --upgrade aura-memory

From the MCP Server section of the README; check the server with `python -m aura mcp ./aura_brain --check`.

Access and checks

How we review

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

  • The Claude Code auto-capture hook stores the conversation verbatim, including any secrets that slip into it.

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

What it does

Aura Memory keeps facts, decisions and code snippets on local disk and serves them to an agent through the `aura-mcp` MCP server. Memory is split into levels, records decay over time and contradictions are tracked. The source is labeled: the user own words rank above web pages, files and tool output, and what a model wrote counts as inference, not the human word. The server offers 11 tools, including `recall`, `store`, `store_decision`, `search` and `maintain`. For Claude Code there is a hook that stores every conversation turn.

Who it is for. People using Claude Code, Cursor or Codex who want an agent to remember project decisions across sessions with no cloud and no fine-tuning.

Good fit when

  • The agent asks about the stack and agreements again every session.
  • You need to tell user facts apart from text found on the web.
  • Several clients must share one local memory.
  • Every conversation turn is saved through a Claude Code hook.

Not a fit when

  • You need cloud memory reachable from many machines with no storage of your own.
  • A notes file in the repository is enough.
  • Memory would hold secrets and encryption and access rights are not set up.

Example request

Call `recall` for this task before answering, then save the decision and its reason with `store_decision`.

MCP

Transport
stdio
Authentication
not required

Environment variables

AURA_PASSWORD secret
Password of an encrypted store, if enabled.

Limitations

It needs `pip install aura-memory` and the Python interpreter that owns the package. The store lives on disk near the project, and an encrypted one takes its password from `AURA_PASSWORD`. The auto-capture hook stores the conversation verbatim, so check that it holds no secrets. The project is large and changes fast, with version 1.64.2 in `mcp.json`. The feature set is large, so `recall` and `store` are enough to start, and the rest can be adopted as needed.

How to disable. Remove the `aura` entry from the client configuration, drop the `hooks` block from `settings.json` if you enabled auto-capture, and run `pip uninstall aura-memory`. Delete the memory directory separately.

FAQ

How do I get the client configuration?

Run `python -m aura mcp ./aura_brain --print-config claude`. Use `cursor`, `vscode` or `generic` instead of `claude`.

How many tools does the server expose?

Eleven: `recall`, `recall_structured`, `store`, `store_code`, `store_decision`, `search`, `insights`, `consolidate`, `get`, `delete` and `maintain`.

How do I store every conversation turn?

Run `python -m aura capture ./aura_brain --print-config` and merge the `hooks` block into `~/.claude/settings.json` or `.claude/settings.json`.

Are embeddings required?

No, local recall works without them and embeddings are optional.

Where is the memory stored?

In the directory you pass to the server as the first argument, for example `./aura_brain`. It can be password encrypted.

Discussion

No questions or reviews yet.

MCP servers
Official

Long-term memory for agents: a remote MCP server, plugins for Claude Code, Cursor and Codex, and an API for your own apps

MCP serverExtended access31.6K

Local memory for agents with a review queue: SQLite, an audit log and a token ledger, one vault for Claude Code, Codex and Cursor

PluginExtended access180
Official

Local persistent memory for coding agents: hooks record every session, and Claude Code, Codex or Cursor recall past decisions through 54 MCP tools

MCP serverExtended access30.3K

iai-memory

iai-memory (Personal Memory Engine)

Official

A local long-term memory MCP server for agents: keeps conversations verbatim and returns the right context at the start of a session

MCP serverExtended access923

Local memory for Claude Code, Codex and Claude Desktop: one Markdown folder and one MCP server shared by all.

MCP serverExtended access103
Official

Long-term memory for agents: new facts supersede old ones, every answer traces to its source, and recall makes no LLM calls

MCP serverExtended access203

README in short

A local cognitive memory for a frozen model: levels, decay, explainability and security profiles. Available as a Python library and as an MCP server.

README badge

Report a problem