SkillFoxx

jevmem MCP

Shared team memory about code: agreements and failed approaches live in git and the agent checks commands against them

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

Extended access

Why this level

  • Message text goes to the external TypeSafe service for scoring and secrets are stripped only by patterns
  • Hooks inspect and may block commands and edits in Claude Code
All reasons and checks

avinash-jetwani/jevmemAuthor:Avinash-jetwani

Install

Manual install

npm install -g jevmem

Installs the CLI, then `jevmem init --tool claude`.

Access and checks

How we review

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

  • Message text goes to the external TypeSafe service for scoring and secrets are stripped only by patterns
  • Hooks inspect and may block commands and edits in Claude Code

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

What it does

jevmem saves decisions, rules and failed approaches from your chats into one `JEVMEM.md` file in the repository. Claude Code, Cursor and Codex read that file, and teammates get it with the code through git. Saving is decided by the Jev model from TypeSafe AI, which answers with a probability whether a remark is worth remembering. A guard check runs in Claude Code. Before a command, edit or file write, jevmem checks the action against the rules and, if it breaks one, stops the call and asks you, showing the rule, for example a ban on hand-editing generated migrations. The `jevmem trust` command makes a rule you wrote block the call instead of asking. In Cursor a plugin with hooks and an MCP server works, in Codex the file is read through MCP, and Claude Desktop connects over MCP. Install with `npm install -g jevmem`, then `jevmem init --tool claude`.

Who it is for. Teams that work with several agents in one repository and want code agreements not to get lost between sessions.

Good fit when

  • The team agreed not to hand-edit `db/migrations/` and the agent in a new session must stop before such an edit
  • In the morning you said we use pnpm, not npm, and a teammate in Cursor gets the rule through git
  • A failed approach to a task should be recorded so the agent does not repeat it next week
  • Work alternates between Claude Code, Cursor and Codex and the rules are one set

Not a fit when

  • You cannot send chat text to an external service: remarks go to TypeSafe for scoring
  • You have no paid TypeSafe API key: the service does not work without it
  • Only the built-in Claude Code memory is needed and a guard with a shared file in git is not

Example request

Remember a rule: migrations in db/migrations come from a generator and are not edited by hand. Then show all saved project rules

MCP

Transport
stdio
Authentication
API key

Limitations

A paid TypeSafe API key that is billed is required, plus Claude Code 2.1.273 or newer for the guard. Message text is sent to TypeSafe for scoring after common secrets are stripped, and there is no telemetry. It is opted into per project. In the author's tests a change a rule forbade stayed in the code in 0 of 15 sessions, checked by hand on version 0.7.7. The guard works in Claude Code, and in Cursor and Codex the file is read as a reference. The Cursor plugin is copied by hand until it is in the marketplace. The license is MIT and the npm package is called `jevmem`.

How to disable. Remove the plugin (`claude plugin uninstall jevmem@jevmem`), the package (`npm uninstall -g jevmem`) and the `JEVMEM.md` file from the repository if the rules are no longer needed.

FAQ

How do I install jevmem for Claude Code?

With `npm install -g jevmem`, `claude plugin marketplace add Avinash-jetwani/jevmem` and `claude plugin install jevmem@jevmem`. A TypeSafe API key and Claude Code 2.1.273 or newer are needed.

What is the guard?

Before a command, edit or file write jevmem checks the action against the rules in `JEVMEM.md` and, if it breaks one, stops the call and asks you.

Where does my chat text go?

To TypeSafe AI for scoring by the Jev model, with secrets stripped first. There is no telemetry and opting in is per project.

How do teammates get the rules?

The `JEVMEM.md` file lives in the repository and arrives with the code through git. In Cursor a plugin reads it and in Codex an MCP server does.

Can a rule be made blocking?

Yes, the `jevmem trust` command turns a rule you wrote from a question into a blocked call.

Discussion

No questions or reviews yet.

MCP servers

compound-memory turns your AI chats into memory about you each night and injects it into new Claude Code and Codex sessions

PluginBroad access113

Local memory for Claude Code, Cursor and Codex in one SQLite file that feeds past decisions and lessons into the agent's context

MCP serverExtended access287

A spreadsheet where a cell holds a value, a formula, a command or an agent, filled in by Claude Code and Codex

CLIBroad access73

A Claude Code plugin that keeps memory across sessions in one local file, with no database, cloud or API keys

PluginExtended access586

A local wiki memory built from what you read: a shared knowledge folder for agents through an MCP bridge and my-curator.

MCP serverExtended access103

A build script and templates that turn an Obsidian vault into an AI's working memory: markdown notes, no vector database

WorkflowExtended access737

README in short

The intro explains that jevmem makes project rules shared across agents and teammates and checks Claude commands before they run. Then come three scenarios: the guard, teamwork through git and a shared file. Next are install steps for Claude Code, Cursor, Codex and Claude Desktop and how Jev scoring works.

README badge

Report a problem