Artificial Memory MCP server
Commit Oct 15, 2026Provides shared project memory, tasks, and rules for AI coding agents to maintain context across different chats and computers.
Also known as notbadov/artificial-memory
- Stars
- 0
- Tools
- —
- Downloads, 30 days
- —
- Latest version
- —
- Last commit
- Oct 15, 2026
- License
- Apache-2.0
What is Artificial Memory MCP?
Artificial Memory provides shared project memory, tasks, and rules for AI coding agents to maintain context across different chats and computers.
Summary generated from the project's README
About Artificial Memory MCP server
Maintaining project context and rules across different AI coding agents is what artificial-memory enables for developers. It provides a shared space for memory, tasks, and project rules so that new chat sessions do not start from zero. This allows users to keep their setup consistent whether they are working on a laptop or a desktop computer.
To use this service, you create an account and connect your machine using the npx artificial-memory connect command, which uses a six-digit code for browser-based confirmation. The system works with AI applications like Codex, Claude Code, Kimi, and OpenCode. It functions by installing specific skills and rules into these apps, which then communicate directly with the remote service over HTTPS.
Security is managed through a launcher that downloads a signed runtime to ensure everything is verified before execution. Users maintain control by approving what an agent proposes to write into the memory, preventing agents from making unverified changes or deploying without permission.
What you can do with it
- 1Syncing context across machinesA developer can maintain the same project rules and connections when switching from a laptop to a work computer.
- 2Preventing repetitive explanationsYou can avoid re-explaining decisions to an AI agent every time a new chat session begins.
- 3Controlling agent autonomyYou can review and approve any information an AI agent attempts to save to the project memory.
- 4Managing agent permissionsA user can review and manage the specific skills and programs that AI agents are allowed to use.
Written from the project's README. Every use case is checked against a quote from it.
README
View on GitHubArtificial Memory is a service: https://artificial-memory.com. This repository holds the part of it that runs on your computer, so that you can read what it does before you run it.
Who it is for
People who work with AI coding apps and are tired of three things:
- every new chat starts from zero, and you explain again what was already decided and why;
- every computer has its own setup: one set of skills and connections on the laptop, another at work;
- an agent can do too much: deploy without asking, or write a guess into memory as a fact.
Three steps to a first result
-
Create an account at https://artificial-memory.com. Your personal space appears with it.
-
Connect a computer. Either install the application for macOS and Windows from https://app.artificial-memory.com/download, or run one command in a terminal (Node.js 18 or newer, nothing else to install):
npx artificial-memory connectThe command shows a six-digit code and opens the browser; you confirm the code there. What the command checks, writes to disk and sends to the service is described step by step at https://app.artificial-memory.com/cli.
-
Work as before. Your AI apps read the memory of the project themselves and propose what to write into it; you approve what stays.
How it works
flowchart LR
subgraph computer["Your computer"]
launcher["npx artificial-memory"] --> client["Synchronization client"]
client -- "skills, rules, connections" --> apps["Codex · Claude Code · Kimi · OpenCode"]
end
client <-- "signed releases, status" --> service[("Artificial Memory")]
apps <-- "memory, tasks, project context (MCP over HTTPS)" --> service
- The launcher downloads a runtime signed with the release key and checks it before it runs anything.
- The client installs into your AI apps what your space approved: skills, rules and the connections to memory, tasks and project context. It changes only what it added itself.
- The apps then talk to the service directly. The client does not send your files, code or chats; what an agent proposes for memory waits for a person to approve it.
Commands
| Command | What it does |
|---|---|
npx artificial-memory connect |
Connect this computer: a six-digit code, confirmed in the browser. |
npx artificial-memory status |
What is connected, the last synchronization, the background schedule. |
npx artificial-memory sync |
Synchronize now. |
npx artificial-memory consents |
Review skills with programs and servers that wait for your approval. |
npx artificial-memory schedule off |
Stop the hourly background synchronization (on starts it again). |
npx artificial-memory disconnect |
End the connection: access is revoked on the server and removed here. |
npx artificial-memory uninstall |
Disconnect and remove everything the client keeps on the computer. |
What is in this repository
The code that runs on your computer:
| Path | What it is |
|---|---|
packages/command-line/ |
The npm package artificial-memory: a launcher with no dependencies. It downloads the signed runtime, verifies it and starts the terminal client. |
adapters/client/ |
The synchronization client and the terminal client. |
adapters/codex/, adapters/claude/, adapters/claude-desktop/, adapters/kimi/, adapters/opencode/ |
What is installed into each AI app: skills, rules, hooks, modes and the connections to memory, tasks and project context. |
adapters/common/ |
What the adapters share. |
tools/ |
The installers of those connections, which the client starts, and the test runner. |
packages/python/agent-platform-core/ |
The modules the installers import. |
tests/ |
The tests of all of the above. They pass in this repository alone: docs/TESTING.md. |
What is not here
- The service itself: the servers that keep memory, tasks and project context, and the web application. A computer talks to them over HTTPS.
- The desktop application.
- Skills and rules. They are not part of this code. A computer receives them in a release
signed with the release key and verifies the signature before anything is installed; the
skills of your own space are written by you in the service. That is why some files here
name a
shared/directory that this repository does not have: it is a directory of a release.
This repository is a generated copy of the code we develop elsewhere: each publication is one commit.
Names you will see in the code
The code is older than the product's name, so a few internal names remain:
- Agent Platform is the earlier name of the client: hence
~/.agent-platform/,agent_platform_coreand theAGENT_PLATFORM_*variables. - Control Plane is the part of the service that hands out releases and receives the status of each computer.
- Desktop Companion is the desktop application.
home-server-2026-09is the id of the release signing key, the one in docs/VERIFYING.md.
Checking what you run
Everything the launcher downloads is signed, and the key is in this repository.
docs/VERIFYING.md shows what exactly is signed and how to check a
download by hand with openssl and Python.
Questions, problems, security
- Questions and bug reports: GitHub Issues.
- A security problem: please report it privately, see SECURITY.md.
Licence
Apache License 2.0, see LICENSE and NOTICE. The runtime the launcher downloads contains software of other authors under their own licences: THIRD_PARTY_NOTICES.md.
Artificial Memory MCP questions
Does the client send my private files or code to the service?
No, the client does not send your files, code, or chats to the service.
Which AI applications are supported?
The service supports Codex, Claude Code, Kimi, and OpenCode.