An MCP-powered health companion that evaluates your wellness context on every AI interaction and injects proactive coaching through MCP tools and Kiro hooks, so you do not have to ask for reminders.
Developers spend 6-10+ hours daily at their desks, leading to:
- π₯€ Dehydration from forgetting to drink water (impairs cognitive function by 15-25%)
- πΊ Musculoskeletal issues from prolonged sitting (back pain, RSI, tech neck)
- π Poor nutrition from skipped meals and excess caffeine
- π§ Mental fatigue & burnout from continuous focus without mindfulness breaks
- ποΈ Eye strain affecting 50-90% of computer workers
These issues directly reduce code quality, problem-solving ability, and long-term health.
DevWell is an MCP server that plugs into your AI coding workflow and adds proactive health coaching, powered by your existing AI subscription. It evaluates your health context on every AI interaction and through Kiro hooks, so you do not have to ask for reminders. Once configured, it:
- π€ Proactive on every interaction - evaluates your health context whenever your AI tool calls it or a Kiro hook fires, then injects coaching without you asking
- π§ Learns over time - tracks which hours and intervention types you respond to. It surfaces correlations once it has enough days of data
- π Tracks analytics - trends, burnout signals, productivity phases and correlations, all computed from your real history
- π― Respects your flow - protects deep focus, only interrupts when critical
- β‘ Zero extra cost - powered by your existing AI tool's subscription or gateway
DevWell runs inside your AI coding tool, not a second terminal. It is built for Kiro and also works in Claude Code and Codex.
DevWell is a Kiro plugin. Its .kiro/ directory is the whole config: .kiro/settings/mcp.json loads the bundled MCP server (34 tools, nothing to build), .kiro/hooks/ make it proactive (a SessionStart health companion and a PostToolUse nudge), .kiro/agents/wellness-coach.md is the coaching voice and .kiro/specs/ are the specs it was built from. Open the repo in Kiro, or point your Kiro MCP config at node /ABSOLUTE/PATH/TO/devwell/plugin/devwell-mcp.cjs.
/plugin marketplace add zkasuran/devwell
/plugin install devwell@devwell
That is it. DevWell is now inside Claude Code:
- 34 MCP tools, so you log and check health by asking (for example "log 2 waters" or "how is my health").
- Slash commands:
/devwell:checkin,/devwell:water,/devwell:break,/devwell:meditate. - Two hooks: a SessionStart hook hands the assistant your health context, a UserPromptSubmit hook injects a nudge only when something is due, so reminders arrive in your flow without you asking.
- The wellness-coach agent for a dedicated coaching voice.
The plugin bundles a self-contained server (plugin/devwell-mcp.cjs), so there is nothing to build after installing. For an always-visible badge, optionally add a status line to your Claude Code settings.json:
{ "statusLine": { "type": "command", "command": "node /ABSOLUTE/PATH/TO/devwell/plugin/devwell-cli.cjs statusline" } }DevWell runs in OpenAI Codex as a plain MCP server too. See the MCP configuration under Quick Start below.
- Node.js 18+ (download)
- One of: Kiro CLI, Claude Code (
claude), or any MCP-compatible tool
# Clone the repository
git clone https://github.com/zkasuran/devwell.git
cd devwell
# Install dependencies
npm install
# Build the project
npm run build
# Verify it works
node dist/cli.js helpAdd to your MCP configuration (.kiro/settings/mcp.json or via Kiro settings):
{
"mcpServers": {
"devwell": {
"command": "node",
"args": ["/absolute/path/to/devwell/dist/index.js"],
"env": {}
}
}
}Add to your Claude Code MCP settings (~/.claude/settings.json):
{
"mcpServers": {
"devwell": {
"command": "node",
"args": ["/absolute/path/to/devwell/dist/index.js"]
}
}
}Add to your Codex MCP configuration:
{
"mcpServers": {
"devwell": {
"command": "node",
"args": ["/absolute/path/to/devwell/dist/index.js"]
}
}
}The point of DevWell is not a separate devwell window you remember to open. It lives inside the tool you already code in, in two ways.
A status line that is always glanceable. devwell statusline prints a one-line health badge:
[π§5/8 | π45m | β0mg | π | π80pts]
Wire it into Claude Code's status line and your hydration, sitting time, caffeine, stress and health score sit at the bottom of the CLI, no extra window.
Hooks that let the assistant nudge you in your flow. devwell nudge prints a short reminder only when something is due, and nothing otherwise, so it never adds noise. devwell context prints the coaching context. Fire them from Claude Code hooks so the assistant raises health in context without you asking.
Ready-to-paste Claude Code config is in integrations/claude-code/settings.json (swap in your absolute path):
{
"statusLine": { "type": "command", "command": "node /ABS/devwell/dist/cli.js statusline" },
"hooks": {
"SessionStart": [{ "hooks": [{ "type": "command", "command": "node /ABS/devwell/dist/cli.js context" }] }],
"UserPromptSubmit": [{ "hooks": [{ "type": "command", "command": "node /ABS/devwell/dist/cli.js nudge" }] }]
}
}On Kiro, the same proactive behavior comes from the agent hooks in .kiro/hooks/ (a SessionStart and PostToolUse health hook plus a UserPromptSubmit monitor). On Codex or any MCP client, the assistant calls the DevWell tools and appends the health badge to its responses.
Once configured, just tell your AI tool:
"Activate DevWell autonomous mode"
That's it. DevWell will:
- Set up your health profile with sensible defaults
- Begin autonomous monitoring immediately
- Learn your patterns over time
- Intervene at optimal moments
- Get better every day
Or customize first:
"Set up my DevWell health profile"
"Configure DevWell autonomous mode - I prefer minimal interruptions"
DevWell evaluates your health on every AI interaction rather than on a background timer. There is no separate daemon: each time your AI tool calls a DevWell tool or a Kiro hook fires, the engine checks your state and can weave in coaching. Once enabled:
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
β You code normally. DevWell handles the rest. β
β β
β Every AI interaction β Autonomous evaluation β
β Health overdue? β AI naturally reminds you β
β Pattern detected? β Strategy adapts automatically β
β Burnout risk? β Proactive intervention β
β Everything fine? β Silent, invisible β
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
Your Activity β Record Pattern β Analyze β Adapt Strategy β Better Intervention
β β
βββββββββ You Follow/Ignore β Feedback Loop ββββββββββββββββββ
DevWell continuously:
- Records when/how it intervenes and whether you respond
- Learns which hours, phases, and styles get the best response from you
- Adjusts intervention timing, frequency, and messaging
- Correlates health behaviors with productivity (water vs focus, breaks vs output) once it has enough days of data
- Predicts optimal behaviors based on your accumulated data
| Phase | Duration | DevWell Behavior |
|---|---|---|
| π’ warming_up | 0-15 min | Silent - let you settle in |
| π΅ building_focus | 15-45 min | Only urgent health items |
| π£ deep_flow | 45-90 min | PROTECTED - critical only |
| π‘ sustained | 90-120 min | Gentle reminders OK |
| π declining | 120-180 min | Proactive suggestions |
| π΄ fatigued | 180+ min | Active intervention |
DevWell uses your existing AI subscription - no extra API costs:
- Kiro: System prompts + hooks inject health coaching context
- Claude Code: a status line health badge plus SessionStart and UserPromptSubmit hooks (see "Run it inside your coding CLI")
- Codex: Tool calls automatically trigger health evaluation
The AI naturally becomes your health companion through context injection.
Multi-factor burnout risk scoring:
- Session duration (how long without proper breaks)
- Break deficiency (missed vs. expected breaks)
- Stress accumulation (tracked over days/weeks)
- Health score decline (pattern degradation)
- Productivity phase (fatigue signals)
When risk is HIGH or CRITICAL, DevWell will strongly encourage you to stop.
- Configurable daily water goal (default: 8 glasses)
- Customizable reminder intervals (default: 60 minutes)
- Visual progress bars with emoji indicators
- Evidence-based hydration tips
- Smart reminders based on time since last drink
- 18 exercises across 8 categories:
- Stretching | Desk exercises | Standing | Walking
- Eye care | Posture | Strength | Light cardio
- Exercises tailored to your work style
- Step-by-step instructions with proper form guidance
- Physical limitation filtering
- 20-20-20 eye care rule integration
- Urgency escalation for prolonged sitting
- 18 food suggestions organized by time of day
- Brain boost ratings (1-5 stars) based on cognitive science
- Meal and snack logging with schedule awareness
- Caffeine tracking with daily limits and cutoff times
- Dietary restriction support (vegetarian, vegan, gluten-free, etc.)
- Cooking willingness consideration (prep time filtering)
- 15 guided meditation techniques across 9 categories:
- Breathing | Body scan | Visualization | Mindful coding
- Micro-meditation | Walking meditation | Focus reset
- Gratitude | Progressive relaxation
- Work-style-aware recommendations
- Stress level tracking with responsive suggestions
- Beginner to advanced difficulty levels
- Developer-specific techniques (e.g., "Code Architecture Visualization")
- Real-time health score (0-100)
- Daily statistics tracking
- Weekly trend reports with improvement suggestions
- Streak tracking for consistency motivation
- Per-category goal progress
- 7 work styles with tailored recommendations
- Complete preference customization
- Persistent profile across sessions
- Working hours awareness
| Tool | Description |
|---|---|
check_health |
Autonomous health snapshot with AI analysis & burnout detection |
autonomous_status |
View autonomous engine status & next planned action |
configure_autonomous |
Configure intervention style, flow protection, quiet hours |
get_coaching_context |
Get AI coaching system prompt (powers autonomous behavior) |
adaptive_insights |
View learned patterns, responsiveness, correlations |
analytics_dashboard |
Full analytics: trends, heatmaps, predictions, milestones |
burnout_assessment |
Detailed burnout risk with contributing factors |
record_health_action |
Train the AI - record if you followed a recommendation |
| Tool | Description |
|---|---|
log_water |
Log water intake (feeds adaptive learning) |
water_status |
View hydration progress |
| Tool | Description |
|---|---|
log_break |
Log break (feeds adaptive learning) |
suggest_exercise |
AI-personalized exercise for your phase & style |
movement_status |
Sitting analytics with autonomous scheduling |
| Tool | Description |
|---|---|
log_meal |
Log meal (system learns patterns) |
log_caffeine |
Track caffeine for sleep analysis |
suggest_nutrition |
AI-powered brain food suggestion |
nutrition_status |
Meal timing & caffeine analytics |
| Tool | Description |
|---|---|
log_meditation |
Log meditation (correlated with stress/productivity) |
suggest_meditation |
AI-selected technique for your phase & stress |
mindfulness_status |
Stress & mindfulness analytics |
set_stress_level |
Report stress (system adapts interventions) |
| Tool | Description |
|---|---|
get_profile |
View profile with autonomous preferences |
set_work_style |
Change work style (adapts all recommendations) |
update_preferences |
Update any health preference |
session_info |
Session with productivity phase detection |
daily_stats |
AI-generated daily insights |
weekly_report |
Trend analysis with coaching |
pause_session / resume_session |
Control autonomous monitoring |
| Tool | Description |
|---|---|
quick |
Ultra-fast health action via shortcodes (w, b, m, c, ok, skip, ...) |
shortcodes |
Show all available quick-action shortcodes |
health_badge |
Compact one-line health status badge to append to any response |
notification_settings |
Configure notification mode, DND hours, frequency, compact mode |
tracking_report |
Event tracking analytics: actions logged, response rates, patterns |
βββββββββββββββββββββββββββββββββββββββββββββββ
β AI Tool (Kiro / Claude / Codex) β
βββββββββββββββββββββββ¬ββββββββββββββββββββββββ
β MCP (stdio)
βββββββββββββββββββββββΌββββββββββββββββββββββββ
β DevWell MCP Server β
β β
β ββββββββββββββββββββββββββββββββββββββββ β
β β Session Manager β β
β ββββ¬βββββββ¬βββββββ¬βββββββ¬ββββββββββββββ β
β β β β β β
β ββββΌβββββββΌβββββββΌβββββββΌβββββ β
β βWaterββMove ββFood ββMeditaβ β
β βTrackββTrackββTrackββtion β β
β βββββββββββββββββββββββββββββ β
β β β β β β
β ββββΌβββββββΌβββββββΌβββββββΌββββββββββββββ β
β β Profile + Stats (Persistence) β β
β βββββββββββββββββββββββββββββββββββββββ β
βββββββββββββββββββββββββββββββββββββββββββββββ
devwell/
βββ .kiro/ # Kiro configuration
β βββ specs/ # Feature specifications
β β βββ devwell-health-mcp.md
β β βββ hydration-module.md
β β βββ movement-module.md
β β βββ nutrition-module.md
β β βββ mindfulness-module.md
β βββ steering/ # Development guidelines
β β βββ project-conventions.md
β β βββ mcp-protocol-guidelines.md
β β βββ health-science.md
β βββ hooks/ # Automated hooks
β βββ pre-commit-lint.json
β βββ test-on-change.json
β βββ health-reminder.json
β βββ build-check.json
βββ src/
β βββ core/ # Core engine
β β βββ types.ts # All TypeScript interfaces
β β βββ session-manager.ts
β β βββ profile-manager.ts
β β βββ stats-manager.ts
β βββ modules/ # Health modules
β β βββ hydration.ts
β β βββ movement.ts
β β βββ nutrition.ts
β β βββ mindfulness.ts
β βββ mcp/ # MCP server layer
β β βββ server.ts
β βββ data/ # Health databases
β β βββ exercises.ts # 18 exercises
β β βββ nutrition.ts # 18 food suggestions
β β βββ meditations.ts # 15 techniques
β βββ utils/
β β βββ paths.ts
β βββ index.ts # MCP server entry
β βββ cli.ts # CLI entry
βββ tests/ # Test suite
βββ package.json
βββ tsconfig.json
βββ README.md
# Run all tests
npm test
# Build and verify
npm run build
# Test CLI directly
node dist/cli.js status
node dist/cli.js water 1
node dist/cli.js exercise
node dist/cli.js meditate
# In-CLI integration commands
node dist/cli.js statusline # one-line health badge for a CLI status line
node dist/cli.js nudge # a proactive reminder, printed only when one is due
node dist/cli.js context # the coaching context for a SessionStart hook| Variable | Description | Default |
|---|---|---|
DEVWELL_DATA_DIR |
Custom data storage path | Platform-specific |
DEVWELL_CONFIG_DIR |
Custom config path | Platform-specific |
| Platform | Data Directory |
|---|---|
| Linux | ~/.local/share/devwell/ |
| macOS | ~/Library/Application Support/devwell/ |
| Windows | %APPDATA%/devwell/ |
Water goal: 1-20 glasses/day
Water reminder: 15-120 minutes
Break interval: 15-120 minutes
Exercise intensity: light / moderate / vigorous
Caffeine limit: 0-600mg/day
Caffeine cutoff: any hour
Meditation duration: 1-30 minutes
Dietary restrictions: vegetarian, vegan, gluten_free, dairy_free, nut_free, low_sugar, high_protein, keto, paleo
Work style: deep_focus, pomodoro, collaborative, creative, debugging, review, learning
DevWell was built entirely using Kiro, leveraging:
-
Spec-Driven Development: All features were first defined as specifications in
.kiro/specs/before implementation. The main spec defined the architecture, module-level specs defined each health domain, and the autonomous engine spec drove the self-improving AI system. -
Steering Files: Development conventions, MCP protocol guidelines, and health science references in
.kiro/steering/ensured consistent code quality, evidence-based recommendations, and proper autonomous behavior. -
Agent Hooks:
SessionStarthook activates autonomous health monitoringUserPromptSubmithook injects health context into every interactionPostFileSavehooks for TypeScript compilation and test runningPostToolUsehook for autonomous health evaluation
-
Custom Agent: The
wellness-coach.mdagent definition provides the AI personality and behavior guidelines for autonomous health coaching. -
MCP Configuration: The
.kiro/settings/mcp.jsonconfigures DevWell as a self-contained MCP server that the AI tool uses for health management. -
Iterative Development: Each system was built incrementally with Kiro - from types β core β modules β autonomous engine β adaptive learning β MCP layer.
- 100% local: All health data stored on your machine only
- No telemetry: Zero data transmitted to any server
- No cloud: Works completely offline after installation
- Your data: Delete
~/.local/share/devwell/to remove all data
MIT License - see LICENSE file.
- Model Context Protocol SDK - MCP server framework
- Health guidelines based on WHO, FDA, American Academy of Ophthalmology, and published research
- Exercise and meditation techniques adapted from established wellness practices
Contributions welcome! Please read the steering files in .kiro/steering/ for code conventions.
Built with πΏ by a developer, for developers. Because great code starts with a healthy coder.