ContextPulse captures your desktop continuously and serves it to any MCP-compatible AI agent. Install in 30 seconds. No cloud, no API keys, no cost.
$ pip install contextpulse-sight
Your AI assistant can refactor your codebase but has no idea what's on your screen. You're the bottleneck.
Snip Tool → paste → wait for AI to process. Every time. For every question. It breaks your flow.
New session, blank slate. Developers spend ~5 hours/week re-explaining context their AI should already have.
Claude Code can't see what Cursor shows. Gemini doesn't know what Copilot just did. Your tools don't share context.
Install what you need. Each package works standalone. Together, they compound.
Runs in your system tray. Captures your screen every 5 seconds with change detection. Serves every frame to any MCP-compatible AI agent via 7 built-in tools.
# In your AI agent (Claude Code, Cursor, etc.)
# Just ask naturally:
"What's on my screen right now?"
# Or via MCP tool call:
get_screenshot(mode="active")
get_screen_text()
get_recent(count=5)Cross-session persistent memory so your AI agents don't start from zero. Decisions, preferences, and project context that accumulates over time and is shared across all your AI tools.
# Agent A learns something at 2pm
memory.store({
"fact": "User prefers pytest over unittest",
"source": "claude-code",
"confidence": 0.95
})
# Agent B knows it at 3pm
memory.recall("testing preferences")
# → "User prefers pytest over unittest"Multi-agent coordination so your AI tools work as a team, not isolated silos. Session protocols, handoff patterns, and shared state between any MCP-compatible agent.
# Morning: Claude Code refactors auth
agent.log("Refactored auth middleware",
files=["auth.py", "middleware.py"])
# Afternoon: Cursor picks up where it left off
agent.context()
# → "auth.py was refactored today by
# claude-code. Review before editing."Most developers don't realize how much time they lose. Find out in 30 seconds.
With ContextPulse: 0 manual screenshots. 1.9 hours/week back.
Get ContextPulse freepip install contextpulse-sight
Python 3.10+. No GPU. No cloud account. No API key.
contextpulse-sight
Starts in your system tray. Auto-captures every 5 seconds. Invisible.
Add the MCP server to Claude Code, Cursor, or any MCP-compatible tool. Your AI can now see your screen.
ContextPulse runs 100% locally. No cloud. No accounts. No telemetry. No data ever leaves your computer.
All captures stay on disk. No network calls. Works offline.
Automatically skip sensitive apps — banking, password managers, private browsing.
Captures stop instantly when you lock your screen or switch users.
Every line of code is auditable. No hidden data collection. No surprises.
| ContextPulse | MCP Screenshot Tools | Screenpipe | |
|---|---|---|---|
| Always-on daemon | ✓ Auto every 5s | ✗ On-demand only | ✓ Continuous |
| Activity search | ✓ FTS5 (titles + OCR) | ✗ | ✓ |
| Time-travel context | ✓ get_context_at() | ✗ | ✗ |
| Smart storage | ✓ 4 modes, -59% disk | ✗ | ✗ Full recording |
| MCP tools | 7 (capture, OCR, search) | 1-2 (capture only) | Add-on |
| CPU / RAM | <1% / <20 MB | 0% (idle) | 5-15% / 200-500 MB |
| Price | Free | Free | $400 |
Sight is free and open source. Memory and Agent packages coming soon.
Under 1% CPU and under 20MB RAM. The daemon uses mss for capture (3ms per frame) and only stores frames that changed. You won't notice it running.
Window blocklist lets you skip specific apps (banking, password managers). Auto-pause on lock screen. Everything stays local -- no cloud, no telemetry.
Any MCP-compatible tool. That includes Claude Code, Cursor, Gemini CLI, and any custom agent using the MCP SDK.
Screenpipe records everything you see, say, and hear -- full video and audio. ContextPulse captures only what your AI agents need: screen context. That's why it uses under 1% CPU vs 5-15%, under 20MB RAM vs 200-500MB. It's also free.
Windows-first, but the capture library (mss) is cross-platform. macOS and Linux support is planned.
Free, open source, and local-first. Your AI agent sees what you see in 30 seconds.