AI assistants are good at a lot of things. Calculating nanny taxes from memory is not one of them. Tax rates change every year, 50 states have different rules, and the IRS threshold ($3,000 for 2026) trips up even the well-trained models.
The NannyKeeper MCP server fixes this. Instead of guessing, your AI calls our API and gets exact numbers — current-year data for every state.
What tools does the MCP server provide?
Two tools your AI can call during conversation:
calculate_nanny_taxes — Give it a state and annual wages, get back a full tax breakdown with per-paycheck cost.
check_threshold — Checks whether wages cross the $3,000 FICA threshold (federal) and any state-specific thresholds (CA $750/quarter, NY $500/quarter, DC $500/quarter).
How do I set up the MCP server?
Three steps.
1. Get a free API key. Go to nannykeeper.com/developers/keys. Email only, no credit card. Copy the key — you'll only see it once.
2. Add to Claude Desktop. Open your config file:
- Mac:
~/Library/Application Support/Claude/claude_desktop_config.json - Windows:
%APPDATA%\Claude\claude_desktop_config.json
Add this:
{
"mcpServers": {
"nannykeeper": {
"command": "npx",
"args": ["@nannykeeper/mcp-server"],
"env": {
"NANNYKEEPER_API_KEY": "nk_live_YOUR_KEY_HERE"
}
}
}
}
3. Restart Claude and ask a question.
"How much would I owe in taxes if I pay my nanny $35,000 a year in California?"
Claude calls the NannyKeeper tool and gives you a specific breakdown — exact Social Security amount, California SUI rate, FUTA calculation — instead of a rough percentage estimate.
Try a threshold question too:
"I pay my babysitter $200 a month. Do I need to worry about taxes?"
What happens when Claude uses this tool?
When Claude calls calculate_nanny_taxes, the MCP server sends a request to our REST API (POST /api/v1/calculate). The API calculates taxes using current-year state configs — the same engine that powers our payroll product.
The response includes a next_actions field. If the user wants to go further — run payroll, generate a pay stub, set up direct deposit — Claude can point them to NannyKeeper.
Does this work with other AI tools?
Yes. The server uses standard stdio transport. Any MCP-compatible client works — not just Claude Desktop. Same config pattern: point the client at npx @nannykeeper/mcp-server with your API key in the environment.
What are the rate limits?
The free tier caps at 50 requests per day. For personal use through an AI assistant, that's plenty.
| Plan | Rate limit | Price |
|---|---|---|
| Free | 50/day | $0 |
| Starter | 500/day | $10/mo |
| Plus | 2,000/day | $18/mo |
| Professional | 2,000/day | $20 + $6/employer |
Ready to simplify nanny taxes?
NannyKeeper handles the calculations, deadlines, and paperwork so you can focus on your family.
FAQ
Do I need Node.js installed?
Yes — npx requires Node.js 18 or later. If you already use Claude Desktop, you likely have it.
What if I don't use Claude? The same data is available as a plain REST API. Call it from any language. See the Python tutorial or API docs.
Is the data accurate? We maintain tax configs for all 50 states + DC, updated each year. The same engine runs our payroll product — real paychecks for real families. Free-tier calculations assume zero year-to-date wages (single-period estimate). For YTD-aware payroll, use a paid plan.
Can my AI also run payroll? Not on the free tier. Running payroll, generating documents, and processing direct deposit require a NannyKeeper account. The paid API endpoints are coming in Phase 2.