Mengram MCP
Memory for AI agents with three types: facts, events and procedures that learn from failures; cloud by key or a local folder
- 207 ★ on the repository
- Last commit
- Apache-2.0
Extended access
Why this level
- In cloud mode memories of your projects and decisions are stored on the mengram.io server
- Hooks run on their own and write to memory after each reply, and a hook may ask for confirmation before a Bash command
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add mcp/mengramDetects 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.
- In cloud mode memories of your projects and decisions are stored on the mengram.io server
- Hooks run on their own and write to memory after each reply, and a hook may ask for confirmation before a Bash command
This is third-party code. Review the repository files before installing.
What it does
Mengram keeps three kinds of memory: semantic (facts and preferences), episodic (events and decisions) and procedural (workflows). Procedures change when they fail. For Claude Code there is a plugin with hooks and an MCP server: the profile loads at session start, including after `/clear` and auto-compaction, a search over past sessions runs before each prompt and after a reply new knowledge is saved in the background. The same hook set installs into Codex and Cursor so the three agents share memory. Cloud mode works with a free key from mengram.io and local mode keeps memory in a Markdown folder with no account: `mengram server --memory ./memory` starts an MCP server for Claude Desktop, Cursor and other clients. Retrieval works across 23 languages, Russian included.
Who it is for. People working in Claude Code, Codex and Cursor who want shared memory that survives context clearing, machine switches and task handoffs.
Good fit when
- The agent starts from zero after `/clear` and the profile and project context should return on their own
- A deploy procedure that keeps failing at one step should change and track successes and failures
- Accumulated Claude Code session history has to be moved with `mengram import claude-code`
- Memory is kept in a local Markdown folder so data stays on the machine and opens in Obsidian
Not a fit when
- Cloud mode is out because of the data: memory goes to the mengram.io service and only the local folder remains
- You only need project documentation search: a dedicated server is simpler
- You do not want background hooks: they install themselves and write to memory after replies
Example request
Recall how we deployed the last release, which steps failed and suggest a procedure that accounts for those failuresMCP
- Transport
- stdio
- Authentication
- key is optional
Environment variables
MENGRAM_API_KEYsecret- Free key from mengram.io for cloud mode, not needed in local mode
Limitations
Cloud mode needs a free key from mengram.io and memory sits on the service's server, with secrets stripped on your machine before history is uploaded. Local mode works with no account but processing needs a key from Anthropic, OpenAI or a local Ollama, and its search matches Unicode words with no embeddings and no query translation. The hooks for Claude Code, Codex and Cursor are installed with `mengram setup`, and Cursor has no per-prompt hook. The comparison table in the README is compiled by the author. The license is Apache-2.0 and the package is `mengram-ai`.
How to disable. Uninstall the plugin (`claude plugin uninstall mengram`) or the hooks with `mengram hook` per the project's guide and delete the key from `~/.mengram/config.json`. The local memory folder is deleted by hand.
FAQ
How do I install Mengram in Claude Code?
Save the key in `~/.mengram/config.json`, run `claude plugin marketplace add alibaizhanov/mengram`, then `claude plugin install mengram@mengram`. Or `pip install mengram-ai && mengram setup`.
What do the hooks do?
At session start they load the profile, before a prompt they search for context, after a reply they save knowledge, before compaction they write the working state down and restore it after.
Can I use it with no account?
Yes, local mode: `mengram local init ./memory`, then `mengram server --memory ./memory`. Memory lives in a Markdown folder.
Does it work with Russian?
Yes, cloud retrieval works across 23 languages and local search matches Unicode words, Russian included, without embeddings.
What is procedural memory?
Workflows with versions and success and failure counts per step. When a procedure fails it evolves.
Discussion
No questions or reviews yet.
Related
MCP serversPiia Engram: portable memory for AI agents
Piia Engram
Local memory about you and your projects: Claude Code, Codex and Cursor read the same preferences and lessons from your disk
Local persistent memory for coding agents: hooks record every session, and Claude Code, Codex or Cursor recall past decisions through 54 MCP tools
Agent memory as a graph of linked facts: decisions and errors outlive the session, with no vector database or paid API calls
Membase keeps one shared memory for AI agents that Claude Code, Cursor and Codex read through MCP, an SDK or the membase command
Engram (getengram.app): cloud memory for agents
Engram (getengram.app)
A remote MCP server that keeps agent conversations verbatim and searches them by meaning across sessions
Engram by Gentleman Programming
Engram (Gentleman Programming)
Persistent memory for AI coding agents in a single Go binary: an MCP server on SQLite with full-text search that remembers your project across sessions
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| Mengram (this card) | no | key is optional | 207 | |
| Mem0 | no | 68,887 | stores facts in a vector database, while Mengram also keeps events and procedures that change after failures | |
| Supermemory | yes | OAuth | 31,602 | works as a cloud service, while Mengram has a local Markdown folder mode with no account |
README in short
The text opens with install commands and an SDK example, then a one-prompt install for any agent, the Claude Code plugin with hooks, shared memory with Codex and Cursor, local folder mode and a comparison table with other memory systems.