AutoMem MCP
Long-term memory for Claude Code, Cursor and Codex: the agent saves decisions and finds them again in later sessions
- 66 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- The agent writes to long-term memory, and what is saved enters the context of later sessions: a wrong entry will keep recurring
- `delete_memory` with tags removes every memory that has any of those tags, with no dry run
- The Claude Code plugin installs bash hooks that fire during the session
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add mcp/automem-mcpDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The agent writes to long-term memory, and what is saved enters the context of later sessions: a wrong entry will keep recurring
- `delete_memory` with tags removes every memory that has any of those tags, with no dry run
- The Claude Code plugin installs bash hooks that fire during the session
This is third-party code. Review the repository files before installing.
What it does
The MCP client connects an agent to the AutoMem service, where memories live in a graph and a vector store. The agent records decisions, preferences and code patterns, and in a later session finds them by meaning, tag or ID. Memories are linked by eleven relationship types, which the agent sets itself. The main tools: `store_memory` saves one memory or a batch of up to 500, `recall_memory` searches by vectors, keywords and tags and expands the result along the graph, `associate_memories` links entries, `update_memory` and `delete_memory` change and remove them, and `check_database_health` checks the service. Claude Code gets a plugin with hooks that remind it about memory at session start. For Cursor, the installer adds the project rules `.cursor/rules/automem.mdc` and a start hook. The AutoMem service is deployed separately: locally with `make dev` at http://localhost:8001 or in the Railway cloud. The `npx @verygoodplugins/mcp-automem install` installer shows a plan of changes and keeps a `.bak` copy of every file it touches.
Who it is for. Developers who work in Claude Code, Cursor or Codex every day and are tired of explaining their code style and past decisions to the agent again.
Good fit when
- The agent forgets project agreements, and you repeat them in every session
- One set of memories is needed in Claude Code at work and in Cursor at home
- Architecture decisions are saved with the links between them and resurface on a similar task
- Before editing code, the agent recalls language and style preferences saved earlier
- Stale memories are cleaned up by selecting a tag and deleting them in one batch
Not a fit when
- There is nowhere to run the AutoMem service: the client does not work without it
- A rules file such as `CLAUDE.md` is enough: for a small project a graph and vectors are overkill
- Memories must not sit in a separate database at all: even a local install keeps them in two databases, `FalkorDB` and `Qdrant`
Example request
Recall which validation library we chose in this project and why, and save the new decision about error handling.MCP
- Transport
- stdio, http
- Authentication
- key is optional
Environment variables
AUTOMEM_API_URLrequired- Address of the AutoMem service, http://127.0.0.1:8001 for a local run.
AUTOMEM_API_KEYsecret- Service key: required for Railway, skipped for a local run.
Limitations
The client works only together with a separate AutoMem service. Locally it is started from the `verygoodplugins/automem` repository; in the Railway cloud the authors estimate the cost after the trial at about 0.5–1 dollars a month.
For Claude Desktop, updating means downloading a new `.mcpb` file: restarting or updating the npm package does not change an installed bundle. `delete_memory` with tags removes every memory that has any of them, with no dry run. The Claude Code hooks are written in bash, so on Windows they need Git Bash, MSYS2 or WSL.
How to disable. The Claude Code plugin is switched off through the `/plugin` command; for other agents, remove the server entry from the configuration. The installer leaves `.bak` copies from which the configuration can be restored.
FAQ
How do I install AutoMem MCP in Claude Code?
Through the plugin: `/plugin marketplace add verygoodplugins/mcp-automem`, then `/plugin install automem@verygoodplugins-mcp-automem`. When you enable it, Claude Code asks for the AutoMem address and key. The service must already be running.
Does AutoMem have to be run separately?
Yes. The client only translates MCP calls into requests to the service API. Locally the service starts with `git clone` and `make dev` and listens on http://localhost:8001. For permanent use the documentation recommends the Railway cloud.
Which agents are supported?
Claude Desktop, Cursor, Claude Code, GitHub Copilot, Codex, OpenClaw, Hermes, Grok Build and Google Antigravity. ChatGPT, Claude.ai and ElevenLabs connect through a separate HTTP proxy, and any other MCP client works with the generic configuration.
Is a key needed?
A local AutoMem needs no key, Railway does. The key and service address are set during install, and the installer writes them into the configuration.
Does the agent decide what to remember?
Yes. The hooks only remind the agent about memory, and in Cursor they write nothing. The model decides during work what to save as durable.
Discussion
No questions or reviews yet.
Related
MCP serversAgent memory as a graph of linked facts: decisions and errors outlive the session, with no vector database or paid API calls
Persistent memory for coding agents: insights are stored as markdown cards and recalled at the start of the next session.
A long-term memory plugin that keeps decisions, preferences and task state across sessions and injects only what a query needs
Membase keeps one shared memory for AI agents that Claude Code, Cursor and Codex read through MCP, an SDK or the membase command
Local memory for agents: stores conversations and project files verbatim and retrieves them with semantic search over MCP
Local memory for an agent: hooks record facts from conversations and the MCP server brings them back in later sessions from one SQLite file
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| AutoMem MCP (this card) | yes | key is optional | 66 | |
| Mem0 | no | 68,514 | another memory service with its own storage scheme and no shared backend with AutoMem | |
| Basic Memory | no | not required | 4,495 | another memory implementation that does not use the AutoMem backend |
| MCP Memory Service | no | not required | 2,090 | a separate memory service for MCP, unrelated to AutoMem |
README in short
The documentation covers installation per platform: Claude Desktop through an `.mcpb` file, Claude Code through the plugin, the rest through the guided installer. A separate section explains remote MCP over HTTP for ChatGPT, Claude.ai and ElevenLabs, and the document list includes memory-rule templates for `CLAUDE.md` and Copilot instructions.