MCP server for live, verified LLM API prices from apipriceindex.com (CC BY 4.0)
Three well-named, read-only tools with clear descriptions (150 - 250 chars each) and complete input schemas. All parameters are typed and described. Output schemas are documented in code comments. Naming follows verb_noun pattern (search_models, get_price, estimate_cost). No critical security issues detected, all tools are read-only, no secrets in parameters. Main gaps: no error recovery guidance in descriptions, no pagination support for search_models (returns up to 20 matches but no cursor/offset), and output schema documentation is in code comments rather than formal schema definitions visible to clients.
Estimate the monthly USD bill of a workload and rank models by cost. Give monthly volumes in millions of tokens (input and output) and an optional share of input served from cache (0-100). Ranks the given model ids — or the whole catalog — from cheapest to most expensive. Cost only: this tool does not rank model quality.
Get the verified price of one model (USD per million tokens). Returns input/output/cached-input prices with verification date, cross-check confidence and source URL. Prices are re-checked daily against official pricing pages. Data: apipriceindex.com, CC BY 4.0.
Search the API Price Index for LLM API endpoints by name, id or provider. Returns matching models with their `id` (use it with get_price / estimate_cost), provider, context window and USD prices per million tokens. Optional filters: provider slug (e.g. 'openai'), minimum context window, open-weights only.
Output schemas not formally documented in tool definitions. Descriptions mention return fields (e.g., 'input/output/cached-input prices') but clients cannot parse structured output schema from the tool definition itself.
search_models lacks pagination parameters (offset, cursor, page). Returns up to 20 matches but no way to fetch beyond that limit or iterate through large result sets.
Error responses in code (e.g., 'Not tracked by the index') lack recovery guidance. Descriptions do not explain what the LLM should do if a model_id is invalid or index is stale.
estimate_cost parameter 'models' is optional but behavior differs significantly (rank entire catalog vs. specific models). Dependency not documented in parameter descriptions.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 66 | 2026-07-28+ | v2 |