@ajinkyakulkarni/cost-calc-mcp

An MCP server that costs AI-agent deployments — wired to the same engine that powers calc.ajinkya.ai, so the numbers it returns are byte-identical to the visual calculator.

The LLM (your Claude Code / Cursor / Claude Desktop) runs the conversation; it never does the arithmetic. Every dollar figure comes from compute_cost, which is hard-gated — it refuses to return a number until the cost-driving inputs (volume, model, hosting, cache rate, per-query token profile) are actually provided. Describe your deployment in plain language; the agent proposes sensible defaults, confirms what it must, computes, and hands back a calc.ajinkya.ai/#w=… link to open the full visual model.

Install

Claude Code:

claude mcp add cost-calc -- npx -y @ajinkyakulkarni/cost-calc-mcp

Cursor / Claude Desktop — add to your mcpServers config:

{
  "mcpServers": {
    "cost-calc": {
      "command": "npx",
      "args": ["-y", "@ajinkyakulkarni/cost-calc-mcp"]
    }
  }
}

Prefer zero install? Use the hosted endpoint

No npm needed — point your client at the hosted Worker:

claude mcp add --transport http cost-calc https://calc.ajinkya.ai/mcp

(Cursor / Desktop: an mcpServers entry with "url": "https://calc.ajinkya.ai/mcp".)

Tools

list_presets · load_preset · get_schema · validate_workload · compute_cost · make_share_link

Plus a cost_interview prompt for a guided session — or just say "help me cost an AI agent."

Why trust the numbers

• Same engine as the website. This package bundles the canonical engine modules verbatim (no fork); a parity check asserts byte-identical output across all 18 presets.
• No hallucinated math. The model only fills in a structured workload and calls the engine.
• Auditable. Every result carries the assumptions it made and the engine's derivation trace.

Requirements

Node ≥ 18. Runs locally over stdio; no API key (the engine is pure math — your client provides the model).

Links

• Calculator: https://calc.ajinkya.ai
• Source: https://github.com/ajinkyakulkarni/ai-cost-calculator-studio

MIT