About okaneland-mcp
@okaneland/mcp is a local stdio MCP server for AI tool pricing and AI product margin calculations. It combines a live, human-verified price index with a local calculator that estimates costs, retained revenue, and break-even points.
Key features
- 01Fetch current pricing for 100+ AI tools
- 02Filter prices by category, query, or ID
- 03Calculate revenue after costs and taxes
- 04Estimate break-even price and usage cap
- 05Run margin calculations locally
Use cases
- 01Compare pricing for AI tools
- 02Estimate an AI product's monthly retained revenue
- 03Set a break-even subscription price
- 04Assess token usage limits for a product
MCP support
The server runs over stdio and exposes pricing lookup and margin calculation tools to an MCP client. Pricing is fetched from a public HTTPS index at call time; margin calculations run locally.
Configuration
Local server config is the same for every client. Copy it into your client’s MCP config file.
Tools
2Saved tool list · updated 3w ago
| Parameter | Type | Required | Description |
|---|---|---|---|
| id | string | No | Exact tool id, e.g. 'creatify' or 'cursor'. |
| limit | integer | No | Maximum number of tools to return. |
| query | string | No | Free-text match against a tool's name, description, id, or category. |
| category | string | No | Exact category name to filter by (see the list in this tool's description). |
Details
From the project README@okaneland/mcp
A small Model Context Protocol server that gives AI agents two things from Okane Land:
| Tool | What it does |
|---|---|
get_ai_tool_prices | Current, human-verified pricing for 100+ AI tools solo builders use, with what's free, what a real month costs, the gotcha, and the last-verified date. Fetched live from the maintained price index. |
calculate_ai_margin | Turns an AI product's MRR into what you actually keep after the token bill, card fees, failed payments, refunds, and tax, plus the break-even price and usage cap. |
It runs on the user's machine over stdio, so there is nothing to host. get_ai_tool_prices fetches the public /api/v1/prices.json at call time, so the data is always as fresh as the site's monthly re-verification.
Install
Add it to your MCP client's config (Claude Desktop, Cursor, Windsurf, and others use the same shape):
{
"mcpServers": {
"okaneland": {
"command": "npx",
"args": ["-y", "@okaneland/mcp"]
}
}
}
Restart the client. The two tools appear in any agent you run there.
Example
You: I want a $20/mo AI writing tool on Claude Sonnet. What do similar tools charge, and would $20 keep money at 500 users?
The agent calls get_ai_tool_prices and calculate_ai_margin, then answers from real data: comparable tools charge more, and at 500 users on a Sonnet-class blend you keep about $3,260/mo (33%), staying profitable down to a $12.25 price or up to 3.3M tokens per user.
Tool reference
get_ai_tool_prices
Inputs (all optional): category, query, id, limit. Returns the matching tool rows plus the index's updated date, method, and source page.
calculate_ai_margin
Inputs: pricePerMonth, payingCustomers, blendedTokenPricePerM, tokensPerUserM, and optional monthlyChurnPct (6.1), intlCardPct (40), vatPct (0), failedPaymentsPct (9), refundsPct (3). Returns gross, tokenBill, cardFees, failedPayments, refunds, tax, keep, keepPct, perCustomerKeep, customersReplacedMonthly, breakEvenPrice, breakEvenTokensPerUserM, and tokenCostExceedsPrice.
Privacy
The server fetches only the public price index over HTTPS and sends nothing about you. The margin calculation runs entirely locally.
Develop
npm install
npm run build # tsc -> dist/
npm test # unit tests pin the margin math to the web calculator
npm run smoke # spins up the server, runs the MCP handshake, calls both tools
The margin math in src/margin.ts is a port of Okane Land's margin calculator; src/margin.test.ts pins its known-good outputs so the two cannot silently drift.
Disclaimer
Numbers are directional, not financial advice.
License
MIT
Comments
New to MCP?
From what the protocol is to installing your first server, these answers cover most first questions.
An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.