RewindDB

RewindDB is an MCP server that interfaces with the local Rewind.ai SQLite database to expose recorded audio transcripts and optical character recognition screen data to AI assistants. Designed for users who run Rewind.ai on their computers to track personal context, this tool enables developers, researchers, and productivity-focused users to integrate their captured screen history and spoken dialogue directly into language model workflows. The system reads encrypted or unencrypted SQLite storage by pointing to the database path and loading connection credentials configured in an environment file. By standardizing access via the Model Context Protocol over standard input and output, it lets external agents inspect what was visually displayed on screen during specific applications or time windows. It also retrieves speech sessions with word-by-word timestamps, isolates user speech from external meeting participants, and returns historical activity patterns. Through this interface, compatible AI assistants can summarize past voice calls, locate documents viewed during earlier browser sessions, extract application usage metrics, and organize conversational records for subsequent fine-tuning or personal knowledge retrieval without manual exports.

Category: AI Memory & Context

Tags: ocr, personal-data, rewind.ai, sqlite, transcripts

Visit RewindDB

How to install and configure RewindDB

1. Clone the repository and install dependencies: bash git clone https://github.com/pedramamini/RewindMCP.git cd RewindMCP pip install . 2. Configure your database path and password by creating a .env file in the project directory or at ~/.rewinddb.env: env DB_PATH=/path/to/your/rewind/database.sqlite3 DB_PASSWORD=your_database_password 3. Register the server in your MCP client configuration (such as Claude Desktop's claude_desktop_config.json): json { "mcpServers": { "rewinddb": { "command": "python", "args": ["/path/to/RewindMCP/mcp_stdio.py"] } } } 4. Restart your MCP client to initialize the server.

What you can do with RewindDB

• Summarize past meetings and calls by retrieving full audio transcript sessions across specific calendar windows or relative time periods like the last hour. - Search optical character recognition records to pinpoint specific text, browser tabs, or code snippets visible on screen during previous work sessions. - Filter recorded audio transcripts to isolate the user's voice from other speakers for compiling clean personal speech datasets. - Inspect daily computer activity patterns, application usage statistics, and active hours to review time spent across individual desktop programs.

Key facts

• https://github.com/pedramamini/RewindMCP
• AI Memory & Context, Databases & Data Stores
• ocr, personal-data, rewind.ai, sqlite, transcripts

Part of MCP Servers

How do I install RewindDB?

You install RewindDB by cloning the official repository with git and running pip install . in a Python 3.6 or newer environment. After installation, create a .env file containing your DB_PATH and DB_PASSWORD parameters, which allows the core library, command-line scripts, and the STDIO MCP server script to access your local database securely.

What can RewindDB do?

RewindDB queries your local Rewind.ai SQLite database to retrieve audio transcripts, screen OCR records, and device activity statistics. Through its MCP tools and command-line scripts, it enables keyword searches across past screen captures, extracts full conversation transcripts with speaker separation, and tracks application active time.

Which MCP clients work with RewindDB?

RewindDB communicates over the standard input and output MCP protocol, making it compatible with any client supporting standard STDIO servers. This includes Claude Desktop, Raycast, Cursor, and custom agent frameworks that invoke local Python processes with appropriate environment variables configured.

Can RewindDB filter transcripts by speaker?

Yes, RewindDB includes functionality to separate speech sources. When querying or exporting transcript data, you can filter for your own voice specifically or isolate external speakers, making it easier to collect targeted speech logs for review or dataset compilation.