For Developers

The problem with deep work

Three Claude sessions running. A build compiling. Pull requests flying. You haven't stood up since morning. In the age of agentic coding, your workflow is more automated than ever — but the human in the loop still needs water, still clenches their jaw, still forgets to breathe.

The irony: the more productive your tools make you, the easier it is to lose yourself in work. There's no natural seam — no commute, no colleague tapping your shoulder, no end-of-meeting stretch. Just you, the terminal, and a body you stopped noticing three hours ago.

Felt as a developer tool

Felt lives in your menu bar and sends gentle body-awareness check-ins throughout the day. But unlike other wellness apps, it's designed for people who live in terminals and talk to AI assistants.

Command-line interface

Full control from your terminal:

# Check your current state
$ felt status
Mode: normal · Phase: work · Last check-in: 47m ago
Today: 6 prompts shown, 4 acknowledged

# Trigger a check-in right now
$ felt trigger
Cue triggered.

# Entering a long meeting? Pause for 90 minutes
$ felt pause --duration 90
Paused for 90 minutes.

# Switch to a gentler mode after a rough morning
$ felt mode recovery
Mode set to recovery.

MCP server for AI assistants

Felt ships an MCP server (Model Context Protocol) that lets AI assistants like Claude interact with your body-awareness practice. Add it to Claude Desktop and your assistant can:

Check when you last acknowledged a body check-in
Trigger a prompt when it notices you've been deep in a task
Adjust your mode based on context ("you mentioned feeling overwhelmed — switching to Recovery mode")
Add custom prompts relevant to what you're working on

Why MCP? The Model Context Protocol lets AI assistants call local tools. Felt's MCP server means your AI assistant isn't just helping you code — it can also notice when the human needs a break. Reads go directly to the local SQLite database; writes go through the localhost API so the app handles state transitions properly.

Localhost API

Everything runs through a local HTTP API on localhost:21323, authenticated with a bearer token. Build your own integrations:

# Get a summary of your day
$ curl -s -H "Authorization: Bearer $(cat ~/.config/felt/token)" \
    http://localhost:21323/api/summary | jq

{
  "mode": "normal",
  "phase": "work",
  "prompts_today": 6,
  "acknowledged": 4,
  "last_checkin_minutes_ago": 47
}

Privacy-first architecture

As a developer, you'll appreciate the architecture: zero network calls, zero telemetry, local SQLite database, all communication via localhost. The HTTP API is loopback-only. There's nothing to firewall, nothing to audit, nothing phoning home. Read the full privacy statement — it's short, because there's nothing to disclose.

Built with

Swift + SwiftUI — native macOS, no Electron
GRDB.swift — SQLite wrapper for persistence
Swift NIO — async HTTP server for the localhost API
swift-argument-parser — CLI framework
MCP swift-sdk — Model Context Protocol integration

All dependencies are MIT or Apache 2.0 licensed. The app runs as an accessory process (no Dock icon, no Cmd-Tab entry) with a clean state machine architecture.

Install

# Download and install (coming soon)
$ brew install --cask felt

# Or install manually
$ make install
Built Felt.app → /Applications/Felt.app
Installed felt CLI → ~/.local/bin/felt
Installed felt-mcp → ~/.local/bin/felt-mcp

Download free trial

← Back to felt.place

Questions about the API or MCP integration? hello@felt.place