Wisp: An Open-Source AI Overlay That Brings MCP to Your Desktop Hotkeys

Developer tooling has long oscillated between full delegation and manual copy-paste. Wisp, a new open-source desktop assistant, stakes out the middle ground: a hotkey-driven overlay that reads your selection, clipboard, app context, browser, documents, or screen snip, then streams an AI response right back where you need it—all without leaving the current window.

Wisp is 100% Python, cross-platform, and permissively licensed. Its extensibility is architectural: addons run in isolated host processes and hook into context, tools, and response events. The standout addon is an MCP bridge that turns Wisp into a Model Context Protocol client—so any MCP server’s tools become available through the same overlay. Wisp also ships a read-only context server (tools include get_selected_text, get_clipboard, get_active_browser_url) that services like Claude Desktop, Cursor, and Codex can consume directly.

For day-to-day use, Wisp offers four core flows:

• Overlay query: Highlight text, press a hotkey, see the answer streamed into a small bubble.
• Vision snip: Ctrl+Alt+Q draws a crop region; the model sees only that image.
• Context-aware rewrite: Captures the app context, rewrites selected text, then pastes exactly where the selection came from.
• Sandboxed agent: Splits a long task across coordinator, builder, and reviewer roles; leaves a report and artifacts.

Providers work through an OpenAI-compatible client, and keys go straight to the OS keychain, not a plain-text config. The project also maintains a list of free-tier and free-credit model endpoints, so the cost to try it can be zero.

While Wisp is actively developed—Linux Wayland and macOS testing are called out as especially welcome—it ships with builds and SHA256 checksums. For anyone who wants an AI co-pilot that stays out of your way and stays open to tinkering, Wisp is worth a close look.

Get started

Grab a prebuilt release from GitHub, verify with SHA256SUMS.txt, or clone the repo and run the launcher:

git clone https://github.com/SunnyLich/Wisp-AI-Assistant.git
cd Wisp-AI-Assistant
./launcher_macos.sh  # or launcher_linux.sh

The first run provisions its own Python environment. For MCP integration, copy the snippet from addons/mcp_bridge/claude_config_snippet.json into your client—Wisp generates the correct interpreter path.