Screen history as a trust layer. Memex captures your screen every 60 seconds, extracts the text, and makes your entire work history searchable. You own the data. It never leaves your machine.
This plugin brings Memex into Claude Code, giving your AI assistant access to what you actually did -- not what you remember doing.
You do real work every day. You write code, read documentation, debug problems, review designs, answer emails. But at the end of the week, what evidence do you have?
For most knowledge workers, the answer is: almost none. Your browser history is a mess. Your git log captures commits, not the four hours of research that led to them. Your Slack messages are scattered across channels you can barely search. The actual record of your labor -- the thing you could point to and say "I did this, here is the proof" -- does not exist.
Memex fixes that.
It is a local-first screen recorder that runs in the background, captures what is on your screen, extracts the text via OCR, and indexes it in a local vector database. When you need to recall what you were working on, search for a link you saw, summarize your week, or prove to a client what you delivered -- it is all there.
Think of it as the vernacular Bible moment for work history. Before the printing press, only institutions controlled the record. Memex puts the record in your hands. Your work history belongs to you, not your employer's Jira board or your client's project management tool.
Two commands:
# 1. Install Memex
curl -fsSL https://raw.githubusercontent.com/joenewbry/memex/main/install.sh | sh
# 2. Start capturing
export PATH="$HOME/.local/bin:$PATH"
memex startOn macOS, grant Screen Recording permission to your terminal app in System Settings > Privacy & Security > Screen Recording.
Once Memex is running, Claude Code gets these tools automatically via the MCP server:
| Tool | What it does |
|---|---|
search-screenshots |
Search your screen history by keyword, with optional date filtering |
daily-summary |
Structured summary of a single day's activity, grouped by time period |
activity-graph |
Timeline showing when you were active, by hour or day |
time-range-summary |
Sampled overview of activity across a date range |
search-recent-relevant |
Semantic search that prioritizes recent, relevant results |
get-stats |
Statistics on your capture database |
start-flow / stop-flow |
Start or stop capture from within Claude |
The plugin includes three slash commands:
/memex-setup-- Install and configure Memex from scratch. Checks dependencies, runs the installer, verifies the MCP connection./memex-capture-- Manage capture: start, stop, check status, view logs, troubleshoot issues./memex-portfolio-- Generate a work proof document from your screen history. Creates a structured markdown report showing what you worked on, when, and with what tools.
You work normally
|
v
Memex captures a screenshot (every 60 seconds)
|
v
OCR extracts all visible text (Apple Vision on macOS, Tesseract on Linux)
|
v
Text is indexed in a local ChromaDB vector database
|
v
Claude Code queries the database via MCP tools
|
v
You ask: "What was I working on yesterday?"
Claude answers with actual data from your screen history.
Everything runs locally. No cloud. No accounts. No telemetry.
"What did I work on this week?" Ask Claude and it will search your screen history, identify the applications you used, the code you wrote, the documents you read, and generate a summary.
"Find that error message I saw yesterday" Search your OCR data for specific text. Better than browser history because it captures everything on screen -- terminal output, IDE content, Slack messages, emails.
"Generate a work report for my client"
Use the /memex-portfolio skill to create a structured proof-of-work document. It pulls from actual screen captures, not self-reported timesheets.
"Help me pick up where I left off" After a break, ask Claude what you were doing. It will find your most recent work sessions and summarize the context.
"Create onboarding documentation" Point Claude at a date range covering a project and ask it to generate docs. It will extract the tools, patterns, and workflows from your actual screen history.
~/.memex/ # Memex installation
refinery/ # Screen capture + OCR engine
mcp-server/ # MCP server (stdio transport)
cli/ # Command-line interface
.venv/ # Python virtual environment
chroma/ # Local ChromaDB data
memex-plugin/ # This plugin (Claude Code integration)
.claude-plugin/plugin.json # Plugin metadata
.mcp.json # MCP server configuration
skills/ # Claude Code slash commands
install.sh # Installer
- macOS 12+ or Linux
- Python 3.10+
- ~500MB disk for the initial install
- Screen Recording permission (macOS)
- Tesseract OCR (Linux only; macOS uses Apple Vision)
Memex is radically local:
- All screenshots are stored on your machine only
- OCR text is indexed in a local ChromaDB database
- The MCP server runs locally via stdio -- no network access
- No data is transmitted anywhere, ever
- No accounts, no analytics, no cloud sync
- You can delete everything by removing
~/.memex
Your screen history is yours. Full stop.
MIT -- see the main Memex repo for details.
- Memex Repository
- Issues
- Contact: joenewbry@gmail.com