# vshulcz/deja-vu [Health: Active]

**Category:** 🧠 Knowledge & Memory  
**Repository:** https://github.com/vshulcz/deja-vu  
**GitHub Stars:** 1320  
**npm Downloads (last month):** 0  
**Views:** 9  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/vshulcz-deja-vu

## Description
Local memory layer over the session histories coding agents already write (Claude Code, Codex CLI, opencode): lexical search, recall tools, session-start auto-recall, secret redaction at index time, cross-machine sync over SSH.

## Tools
Capabilities this server exposes over MCP:

- **deja** — Depends on the mode, below.

## Claude Desktop Quick Installation
Install path detected from listing signals. Uses `npx` (confidence: high):

```json
"mcpServers": {
  "deja-vu": {
    "command": "npx",
    "args": ["-y","@vshulcz/deja-vu"]
  }
}
```

## Documentation

## What vshulcz/deja-vu MCP server does

vshulcz/deja-vu turns coding-agent session history stored on disk into a searchable local memory layer. It can index sessions from Claude Code, Codex CLI, opencode, Cursor, and other supported agents, including history created before installation. The shared index lets one agent retrieve work performed in another agent, so decisions and previous fixes can follow a project across tools.

The indexed material includes conversation content as well as work context such as files opened during turns, commands and their exit statuses, and exact spans replaced by edits. Secrets and tokens are removed while the index is built. The system is local and does not require a model or hosted server; optional semantic recall can be enabled separately.

## How it works

Running `deja install --auto` discovers supported agents, connects MCP recall where applicable, enables session-start recall where supported, and creates the initial index. Later sessions can receive relevant history when they open, and recall can also occur after a command fails. The MCP interface exposes a `deja` tool whose behavior depends on the selected mode.

The terminal interface searches across indexed sessions. A plain `deja` command opens the interactive search view, while `deja brief` or `DEJA_TUI=0` produces a text summary. Searches can be narrowed by agent, relative date, or project. Results can be ordered by relevance or date, and a selected session can be resumed in its original agent or handed off to another one.

The index is shared across projects and agents on the same machine. `deja sync ssh laptop` transfers it between machines over SSH, without an intermediate cloud service. Sessions deleted by an agent remain available in a deleted view until explicitly removed or restored.

## Setup and configuration

On macOS or Linux, install with:

```sh
curl -fsSL https://raw.githubusercontent.com/vshulcz/deja-vu/main/install.sh | sh
deja install --auto
```

The installer places the binary in `~/.local/bin`. If that directory is not already on `PATH`, use the path printed by the installer or start a new shell after updating `PATH`.

Other installation paths include Homebrew, Scoop, winget, npm, and Go. The documented Go command is:

```sh
go install github.com/vshulcz/deja-vu/cmd/deja@latest
```

The project supports macOS, Linux, and Windows. Set `DEJA_TUI=0` to disable the interactive terminal interface and print text output instead.

## Tools and capabilities

- Search indexed session history with lexical queries.
- Recall earlier decisions, fixes, and project context through the `deja` MCP tool.
- Start sessions with automatically retrieved history where the agent supports it.
- Search by agent, date range, or project and sort by relevance or date.
- Resume a session in its original agent or continue it in another agent.
- Mark decisions as accepted or rejected with `deja promote`.
- Delete or restore indexed sessions.
- Synchronize history between machines over SSH.
- Generate terminal or SVG statistics cards with `deja stats --card`.

## Limitations and notes

Automatic recall depends on support from the agent being configured. The server works from session histories already saved on disk, so its available memory depends on what those agents have written and what has been indexed. Semantic recall is optional rather than required. The default installation flow is documented for macOS and Linux, while the binary is also available for Windows through the listed installation methods.

_Full upstream README: https://allmcps.com/mcp/vshulcz-deja-vu/readme_