ModelCostComparison MCP
Measured by MCPmetrics:
Connect your AI client to the same pricing source and calculation engine used by ModelCostComparison.
Install in Cursor
Connect
Remote endpoint: https://modelcostcomparison.com/api/v1/mcp
Transport: Streamable HTTP. No API key or local package required.
Cursor
Use the install link above and confirm the server in Cursor. Alternatively, merge the contents of mcp.json into your Cursor MCP configuration, preserving existing servers. Enable ModelCostComparison in the agent's tools.
Claude
Open Connectors, choose Add custom connector, name it ModelCostComparison and paste the endpoint above. Connect it and enable its tools for your conversation. Organization settings may require an administrator to add it.
Other clients
Add a remote MCP server with the endpoint above and Streamable HTTP transport.
Tools
| Tool | Purpose |
|---|---|
list_models | Find model and exact tariff identifiers. |
get_model_pricing | Get rates, sources and review dates. |
calculate_cost | Calculate input, output and cached-input token costs. |
compare_model_costs | Compare offers for the same token workload. |
find_cheapest_models | Rank eligible reviewed offers for a token workload. |
list_providers | Summarize providers, offers and current price ranges. |
Try it
- “Use ModelCostComparison to list available OpenAI models and show pricing sources and review dates.”
- “Find reviewed offers for OpenAI, Anthropic and Google. Compare 1 million input tokens and 200,000 output tokens, with no cache and at most 8,000 input tokens per request.”
- “For the selected offer, estimate 1 million input tokens including 500,000 cached input tokens, plus 100,000 output tokens. Maximum input per request is 8,000 tokens.”
Start with model discovery; use the exact offer IDs returned by the service. Missing or ambiguous workload details should be clarified before calculating.
Scope
Results are estimates of text-token subtotals in USD, not invoices or model-quality rankings. Cache writes, storage, tools, search and taxes are excluded. Expired or unreviewed offers cannot be calculated. Unsupported offers are reported separately in comparisons. Pricing is served from the website's reviewed deployment snapshot, not fetched from providers on every request.
The service is rate limited. On HTTP 429, wait for Retry-After before retrying. HTTP 503 indicates temporary unavailability. No provider credentials are needed, and tools do not call provider models.
This repository contains connection settings and documentation only. Pricing and calculations remain on the hosted service.
References
Official MCP Registry
Published as io.github.Lazige/modelcostcomparison, version 1.0.1. Registry entry. The registry metadata lives in server.json.