MCP Server

Your pricing model,
accessible to AI

CostPad runs a local Model Context Protocol server, so Claude, Cursor, and other MCP clients can read your model, run scenarios, and export reports — without leaving their flow.

CostPad MCP — example session
Claude Codeget_simulation→ 4 tiers, 610 users, 72.4% margin
Claude Coderun_scenario→ Pro at $59/mo → margin 76.2%
Cursor IDEget_economics→ COGS/user $4.12, ARPU $38.50
Claude Codeexport_report→ summary written to clipboard

Illustrative session — tool calls and results depend on your own model.

Setup

Three steps to connect

The desktop app runs the server for you. To run it standalone, install it and point your MCP client at it.

1
Install
Run the server with npx — no global install needed.
$ npx @costpad/mcp@latest
2
Configure your client
Add CostPad to your MCP client config (Claude Desktop shown).
{
  "mcpServers": {
    "costpad": {
      "command": "npx",
      "args": ["@costpad/mcp"]
    }
  }
}
3
Ask away
Restart the client and your model is available. Ask it to read your margins or test a price.
"What's my blended margin at $59 Pro?"
The MCP server is published to npm as @costpad/mcp.
Tools

What the server exposes

A focused set of tools — read your model, change it, or compute against it.

read
get_simulation
Return the current model — tiers, costs, mix, and headline margin.
compute
get_economics
Blended unit economics: COGS/user, fees/sub, net revenue, margin.
write
update_cogs
Add, edit, or remove cost items across all three cost types.
write
update_tiers
Change tier prices, included usage, and naming.
write
update_mix
Set how many users sit in each tier.
compute
run_scenario
Run a what-if with overridden values and return the result.
compute
get_growth
Project MRR, break-even month, and payback from growth inputs.
read
export_report
Generate a shareable summary of the current model.
Transports & clients

Works with your stack

Standard MCP transports, so any compliant client can connect.

Claude DesktopClaude CodeCursorWindsurf
stdio transport
The default for local clients like Claude Desktop and Cursor. The server runs as a child process — nothing leaves your machine.
SSE transport
For clients that connect over HTTP. Useful when the server runs separately from the client.

Bring your numbers into your AI flow

The MCP server ships with the desktop app, or run it standalone via npm. Free to run, public on npm.