Kira

One MCP. Your agent becomes a genius.

Stop managing CLAUDE.md files, .cursorrules, and skill folders across projects. Install Kira once — your AI agent automatically finds the right instructions, avoids known mistakes, and executes flawlessly.

Privacy by design. Kira learns from agent outcomes via opt-in telemetry that redacts secrets, paths, and identifiers locally before write AND server-side before storage. Run npm run demo:privacy to see exactly what leaves your machine. Full wire format and opt-out in PRIVACY.md.

Install (10 seconds)

Add this snippet to your MCP host config:

{
  "mcpServers": {
    "kira": {
      "command": "npx",
      "args": ["kira-mcp"]
    }
  }
}

That's it. Your agent now has Kira.

The snippet above works as-is in every one of them — just paste it under mcpServers (or the equivalent key for your client).

Demo

What happens

Before Kira: Agent deploys to Vercel, forgets env vars, app crashes. Retries 3 times. Burns tokens.

After Kira: Agent automatically calls kira_lookup("deploy vercel") before acting. Gets step-by-step instructions + a Scar warning: "847 agents forgot env vars — run vercel env ls first." Deploys correctly on the first try.

Three tools, zero config

Auto-firing

You don't call Kira. Kira tells your agent to call it. Via MCP instructions, the agent automatically looks up skills before starting any task. You literally do nothing.

What's inside

31 Skills (and growing daily)

12 Scars (past failure patterns)

Scars warn your agent about mistakes other agents already made:

• Next.js "use client" directive missing — client hooks in server components

• Vercel deploy succeeds but app crashes — missing env vars

• Stripe webhook signature fails — body already parsed

• Auth.js signIn/signOut wrong import — server/client mixup

• Prisma types are stale — forgot generate

• Clerk middleware in wrong directory — auth silently broken

• Supabase RLS not enabled — data publicly exposed

• Tailwind v4 PostCSS config wrong — v3 plugin breaks v4

• Vitest path alias mismatch — tsconfig vs vitest.config desync

• And more — hit counts updated from real agent data

7 Routes (goal-to-plan)

Ask "build a web app" → Kira returns 8 ordered steps, each with its skill and scars:

1. Tailwind CSS v4
2. shadcn/ui
3. ESLint
4. Prisma + Scar: "don't forget prisma generate"
5. Clerk Auth + Scar: "middleware goes in root, not app/"
6. Vitest
7. GitHub Actions CI
8. Vercel Deploy + Scar: "check env vars before --prod"

How it works

Your agent gets a task
    ↓
Kira auto-fires (MCP instructions)
    ↓
kira_lookup("deploy vercel", context: ["nextjs"])
    ↓
Returns: Skill (step-by-step) + Scars (what to avoid)
    ↓
Agent announces choice → follows instructions → reports result
    ↓
kira_report("community.deploy-vercel-nextjs.v1", "success")

Skills are natural language Markdown — no executable code, no injection risk.

Why not just use CLAUDE.md?

Telemetry

Kira sends anonymous outcome data to a central Worker so the community can improve Skills and surface new Scars.

Full schema, redaction rules, retention, and opt-out instructions: PRIVACY.md.

Contributing

The first 1,000 contributors get permanent free access to all Kira features (including future Pro tier).

See CONTRIBUTING.md for how to add Skills and Scars.

Links

• npm

• Design Philosophy

• Business Plan

• Usage Guide

Where agents shine.

A B Button Corporation project.