ComparEdge LLM Cost

What the ComparEdge LLM Cost MCP server does

ComparEdge LLM Cost MCP server gives an MCP-compatible agent current pricing data and local cost calculations for 69 models from 17 providers. It works with input tokens, output tokens, call counts, context limits, provider names, and pricing tiers. Results can cover one request, repeated calls, or longer-term usage projections.

The server is intended for planning and comparison rather than making provider API requests. It can show the estimated price for a workload, rank several models by the same token pattern, identify inexpensive models, and project daily, monthly, or yearly spending.

How it works

Pricing comes from the public llms-model-prices.json feed. The feed contains USD rates per one million tokens and is populated from provider pricing checks that include verification dates. The server refreshes its local copy on a six-hour interval. If refreshing fails, it continues using the last successful copy.

All calculations happen inside the server. It does not call model-provider APIs, so using it does not require provider credentials or expose an API key. Model references can use several forms, such as a provider-qualified name or a readable model label. If the resolver cannot identify one model confidently, it returns possible matches rather than silently selecting one.

Setup and configuration

Run the server through npx with the package configuration below:

{
  "mcpServers": {
    "llm-cost": {
      "command": "npx",
      "args": ["-y", "@comparedge/llm-cost-mcp@latest"]
    }
  }
}

No environment variables, account, or API key are required. Claude Desktop uses ~/Library/Application Support/Claude/claude_desktop_config.json; Cursor adds MCP servers from its Settings interface; and VS Code with Copilot reads .vscode/mcp.json. Restart the client after adding the server so its tools become available. The implementation uses JSON-RPC 2.0 over standard MCP stdio.

Tools and capabilities

The six server tools are:

• estimate_cost: calculate a single-call or repeated-call estimate from a model and token counts.
• compare_models_cost: price the same input/output workload across two to six models.
• monthly_budget: estimate daily, monthly, and annual spending for a recurring workload.
• cheapest_models: find lower-cost models, optionally requiring a minimum context size.
• list_models: inspect tracked models, rates, context information, and tiers, optionally filtered by provider.
• list_providers: list providers, model totals, and each provider's cheapest selection.

The server also exposes prompts for describing a workflow, selecting a low-cost model for a task, and forecasting a budget from expected usage. cheapest_models uses a blended input/output weighting of three to one, so its ranking is a screening step rather than a substitute for pricing the actual token split.

Limitations and notes

ComparEdge LLM Cost MCP server provides estimates based on the tracked feed and the token counts supplied by the user. It does not invoke provider APIs or measure actual usage. Estimates are therefore dependent on choosing the correct model and entering representative input and output volumes.

The catalog covers the models and providers present in the feed, rather than every available model. Cached or batch pricing is shown only when the relevant model supports those options. For workloads with uncertain token counts, the included guidance is approximate; confirm the result with measured usage when accuracy matters. The catalog data is available under CC BY 4.0, while the server is MIT licensed.