memov

MemoV is a memory layer for AI agents that uses MCP to provide a seamless, Git-backed history of prompts, context, and diffs. The memov MCP server enables agents to trace and version AI interactions during a session by launching a memov-mcp-launcher adapter via uvx. Once configured in your editor or tooling, you can start an MCP session that automatically records interactions, branches, and snapshots, while offering a UI for viewing history and diffs. The tooling supports workflows like one-click MCP integration for Claude Code, Codex, or other agents, with semantic search and replay capabilities through the MemoV timeline.

With MCP, you can add Memov as a server in your project workspace (for example via VS Code or Cursor integrations) and then invoke the memov-mcp-launcher through the uvx pathway. This enables automatic versioning of prompts and context alongside code changes, while preserving a clean Git history in your project. The UI at the memov web interface (accessible at the configured URL) lets you filter branches, view diffs, and jump to specific snapshots, making it easy to trace how prompts and context influenced your AI-assisted development sessions.

Prerequisites

• Install uv (Python-based universal installer) as described below
• Git installed on your system
• Optional: curl or wget for quick install scripts

Install UV

• macOS / Linux: curl -LsSf https://astral.sh/uv/install.sh | sh
• Windows: powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"

Verify installations

• uv --version
• git --version

Set up MemoV MCP server in your project

1. In your project root, install the MemoV MCP launcher via uvx from the MemoV repository:

# Example for installing the MCP launcher via uvx from MemoV's repo
uvx --from git+https://github.com/memovai/memov.git memov-mcp-launcher stdio ${PWD}

2. Add the MCP server configuration to your workspace (example for VS Code or Cursor):

• VS Code (settings.json) or your MCP config file:

{
  "mcpServers": {
    "memov": {
      "command": "uvx",
      "args": [
        "--from",
        "git+https://github.com/memovai/memov.git",
        "memov-mcp-launcher",
        "stdio",
        "${workspaceFolder}"
      ]
    }
  }
}

3. Start using MemoV MCP through your editor commands or the UI. For example, invoke the MCP client to connect to the memov server named memov, and begin a session to capture prompts, context, and code diffs.

4. (Optional) If you prefer the original command line flow, you can run the launcher directly via uvx with the appropriate from-URL and workspace path, as shown in the earlier step.

Tips and caveats:

• The memov MCP integrates as an uvx-based launcher; ensure the memov-mcp-launcher component exists at the repository path configured by the --from argument.
• In some editors (e.g., Antigravity mode), you may need to provide an absolute path to your project instead of ${workspaceFolder}.
• If you enable VectorDB (RAG) mode for semantic search, install the [rag] extras when cloning or configuring memov to gain enhanced validation and debugging tooling.
• The MemoV UI is exposed at http://localhost:38888 by default; ensure your environment allows local network access to that port.
• For teams, consider adding a .memignore to exclude large or sensitive directories from the MCP history.
• If you encounter issues with MCP path variables, replace ${workspaceFolder} with the explicit absolute path to your project during configuration.