# deja-vu — Coding Agent Memory & Session Search

> deja-vu — Coding Agent Memory & Session Search is an open-source MCP server for knowledge and memory. Local search and shared memory across your coding agents' session history. It runs locally on your machine and is started by your AI client.

- **Kind:** MCP server
- **Pricing:** Open source
- **Platforms:** CLI, macOS, Linux, Windows, Web, API
- **Website:** https://vshulcz.github.io/deja-vu/
- **Repository:** https://github.com/vshulcz/deja-vu
- **MCP Registry name:** io.github.vshulcz/claude-codex-cursor-agent-memory-deja-vu
- **Categories:** [knowledge and memory](https://ai-powered.com/categories/knowledge-and-memory), [AI models and LLM tooling](https://ai-powered.com/categories/ai-and-llm), [agent tooling](https://ai-powered.com/categories/agent-tooling)
- **Listing updated:** 2026-11-20

## About

The README says deja indexes the transcripts that coding agents already write to disk, such as ~/.claude, ~/.codex and Cursor's state database, and builds a local inverted index from them. Because it reads existing history in place, a session started months before installation can still be found, and no capture step has to be run ahead of time.

The index is served through a terminal search screen opened by running deja with no arguments, a CLI, a browser view generated by deja view, and an MCP server exposing a single deja tool with several modes such as recall, blame, fix and handoff. Lifecycle hooks can inject recalled context at session start, before edits and after failures. Credentials are replaced at index time with redaction markers, and the README states that indexing and search run locally with no remote inference.

Beyond search, it can hand a task from one agent to another, recover uncommitted edit spans that an agent overwrote, and map commits or lines back to the sessions that produced them. Optional semantic reranking works only if the user points it at a local or OpenAI-compatible embeddings endpoint.

## Features

- **Cross-agent session search:** Searches the stored transcripts of many coding agents from one shared local index.
- **Existing history indexing:** Reads transcripts already on disk, so sessions from before installation are searchable.
- **Agent handoff:** Transfers goal, constraints, touched files and open work to another agent.
- **Uncommitted code recovery:** Restores edit spans that an agent replaced before they were committed to git.
- **Local-only indexing:** Indexing and search run on the machine, with credentials stripped at index time.
- **Compaction recovery:** Injects a bounded recovery packet after context compaction or resets.

## Good for

- Finding how a bug was fixed in an earlier session with a different coding agent
- Resuming an interrupted task in another agent without rebuilding context
- Recovering code that an agent overwrote before it was committed
- Tracing which past agent session wrote or changed a given line

## Install

### Claude Code (Terminal)

```sh
claude mcp add claude-codex-cursor-agent-memory-deja-vu -- npx -y https://github.com/vshulcz/deja-vu/releases/download/v0.25.0/deja-vu_0.25.0_registry.mcpb
```

### Claude Desktop (claude_desktop_config.json)

```json
{
  "mcpServers": {
    "claude-codex-cursor-agent-memory-deja-vu": {
      "command": "npx",
      "args": [
        "-y",
        "https://github.com/vshulcz/deja-vu/releases/download/v0.25.0/deja-vu_0.25.0_registry.mcpb"
      ]
    }
  }
}
```

### Cursor (~/.cursor/mcp.json)

```json
{
  "mcpServers": {
    "claude-codex-cursor-agent-memory-deja-vu": {
      "command": "npx",
      "args": [
        "-y",
        "https://github.com/vshulcz/deja-vu/releases/download/v0.25.0/deja-vu_0.25.0_registry.mcpb"
      ]
    }
  }
}
```

### VS Code (.vscode/mcp.json)

```json
{
  "servers": {
    "claude-codex-cursor-agent-memory-deja-vu": {
      "type": "stdio",
      "command": "npx",
      "args": [
        "-y",
        "https://github.com/vshulcz/deja-vu/releases/download/v0.25.0/deja-vu_0.25.0_registry.mcpb"
      ]
    }
  }
}
```

### Windsurf (~/.codeium/windsurf/mcp_config.json)

```json
{
  "mcpServers": {
    "claude-codex-cursor-agent-memory-deja-vu": {
      "command": "npx",
      "args": [
        "-y",
        "https://github.com/vshulcz/deja-vu/releases/download/v0.25.0/deja-vu_0.25.0_registry.mcpb"
      ]
    }
  }
}
```

### Other clients (your client's MCP settings (most use this shape))

```json
{
  "mcpServers": {
    "claude-codex-cursor-agent-memory-deja-vu": {
      "command": "npx",
      "args": [
        "-y",
        "https://github.com/vshulcz/deja-vu/releases/download/v0.25.0/deja-vu_0.25.0_registry.mcpb"
      ]
    }
  }
}
```

## FAQ

### Does deja send session data to an LLM?

The README says indexing and search run locally with no remote inference, and network requests occur only for explicit update, sync, version checks and configured embedding endpoints.

### Which install methods are listed?

The README lists an install script for macOS and Linux, npx, Homebrew, Nix, go install, and Scoop or winget on Windows, plus a signed .mcpb bundle for desktop MCP clients.

### Is semantic search required?

No. The README says lexical retrieval and MCP recall work normally when no embeddings endpoint is configured or reachable.

### Is deja-vu — Coding Agent Memory & Session Search free?

Yes. deja-vu — Coding Agent Memory & Session Search is open source, so you can use and self-host it for free.

### How do I add deja-vu — Coding Agent Memory & Session Search to Claude Code?

Run this command in your terminal: claude mcp add claude-codex-cursor-agent-memory-deja-vu -- npx -y https://github.com/vshulcz/deja-vu/releases/download/v0.25.0/deja-vu_0.25.0_registry.mcpb

### What are the best alternatives to deja-vu — Coding Agent Memory & Session Search?

The closest alternatives to deja-vu — Coding Agent Memory & Session Search are Deja Vu, Agent Memory MCP, Cross-Agent Memory, CogniCore and Seamless.

## Where this comes from

- Description: written from the project's README, the project's website and its MCP Registry entry.
- Install steps: from the official MCP Registry entry.

## Alternatives

- [Deja Vu](https://ai-powered.com/mcp/deja-vu): Deja Vu is a free MCP server for developer tools.
- [Agent Memory MCP](https://ai-powered.com/mcp/agent-memory-mcp): Agent Memory MCP is an open-source MCP server for knowledge and memory.
- [Cross-Agent Memory](https://ai-powered.com/mcp/cross-agent-memory): Cross-Agent Memory is an open-source MCP server for knowledge and memory.
- [CogniCore](https://ai-powered.com/mcp/cognicore): CogniCore is an open-source MCP server for knowledge and memory.
- [Seamless](https://ai-powered.com/mcp/seamless): Seamless is an open-source MCP server for agent tooling.
- [Memoria](https://ai-powered.com/mcp/memoria-2): Memoria is an open-source MCP server for knowledge and memory.
- [Coders Talk](https://ai-powered.com/mcp/coders-talk): Coders Talk is a freemium MCP server for developer tools.
- [codeagent-memory](https://ai-powered.com/mcp/codeagent-memory): codeagent-memory is an open-source MCP server for knowledge and memory.

---
Source: https://ai-powered.com/mcp/deja-vu-coding-agent-memory-and-session-search
