CostInflation

Explore and compare prices

Explore category price histories, compare measured quantities across supported U.S. areas, and assemble research with canonical CostInflation citations.

What you can ask

Compare ground coffee and egg price changes in Houston and Los Angeles; show a table with dates.
Example path ·

MCP connector

Server URL: https://mcp.costinflation.com/mcp
Login: None
Protocol: MCP 2026-07-28
Instructions: CostInflation supplies composable evidence and exact calculation tools; the host agent owns the user's intent, interpretation and recommendations. Use or search to resolve public category IDs and canonical units; fetch or supplies supported area IDs, observation coverage and citable evidence. Search and fetch are data-first preparation: retrieve what the task needs without narrating each lookup or creating a view for every category. Use for dated series, for multi-category area comparisons, for two boundaries and for daily statistics. Use for current explicit quantities, for their dated totals and for fixed-quantity contributions. normalizes supplied quotes; and expose published bands and matched package evidence with their methods. New calculation tools are data-first: pass returned view_data unchanged to with line, bars, contribution, matrix or table when inspection helps. The original history and basket tools also supply views automatically. Use complete structuredContent to answer the actual question. response_summary is a compact factual preview, not a prescribed answer or recommendation; preserve its essential coverage warnings. Match explanation depth to the user's question and cite returned evidence. Use charts or tables when they help inspection; avoid duplicating the interactive view's full tables as prose. For clients without a view, give the relevant facts and a task-relevant table when useful. Keep complete relevant evidence available to agents; never dump raw tool JSON into a human answer. supplies dated observations. requires an explicit unit matching the user's quantity: five pounds is quantity 5 with unit pounds, not five display_units. display_units counts whole displayed package equivalents, so use it only for an explicitly requested number of those equivalents. Use server-calculated changes and complete_area_ids_by_total for ascending basket comparisons; the order uses unrounded totals and excludes partials. Retain stated ties. Do not rank partial subtotals as full baskets or claim checkout savings. Keep actual dates, quantities, area scope and canonical URLs in findings. Use only URLs returned by the tools for CostInflation hyperlinks; use links.methodology_url for general methodology and the returned category URL for category-specific methods. Never invent a path or append a guessed path or fragment. Ask for clarification when categories or units are ambiguous. Do not infer missing history, causality or national coverage. Keep daily history statistics distinct from underlying product distributions. Package matches cover consecutive days with unchanged identity/title/brand; describe the matched cohort and avoid inferring shrinkflation or intent. Never sum marginal percentiles into basket percentiles.

Tools

Reads data15

Skills

1 in the ChatGPT listing

What you can ask

Compare ground coffee and egg price changes in Houston and Los Angeles; show a table with dates.

Example path ·

MCP connector

Server URL

https://mcp.costinflation.com/mcp

None

Protocol

MCP 2026-07-28

Instructions

CostInflation supplies composable evidence and exact calculation tools; the host agent owns the user's intent, interpretation and recommendations. Use or search to resolve public category IDs and canonical units; fetch or supplies supported area IDs, observation coverage and citable evidence. Search and fetch are data-first preparation: retrieve what the task needs without narrating each lookup or creating a view for every category. Use for dated series, for multi-category area comparisons, for two boundaries and for daily statistics. Use for current explicit quantities, for their dated totals and for fixed-quantity contributions. normalizes supplied quotes; and expose published bands and matched package evidence with their methods. New calculation tools are data-first: pass returned view_data unchanged to with line, bars, contribution, matrix or table when inspection helps. The original history and basket tools also supply views automatically. Use complete structuredContent to answer the actual question. response_summary is a compact factual preview, not a prescribed answer or recommendation; preserve its essential coverage warnings. Match explanation depth to the user's question and cite returned evidence. Use charts or tables when they help inspection; avoid duplicating the interactive view's full tables as prose. For clients without a view, give the relevant facts and a task-relevant table when useful. Keep complete relevant evidence available to agents; never dump raw tool JSON into a human answer. supplies dated observations. requires an explicit unit matching the user's quantity: five pounds is quantity 5 with unit pounds, not five display_units. display_units counts whole displayed package equivalents, so use it only for an explicitly requested number of those equivalents. Use server-calculated changes and complete_area_ids_by_total for ascending basket comparisons; the order uses unrounded totals and excludes partials. Retain stated ties. Do not rank partial subtotals as full baskets or claim checkout savings. Keep actual dates, quantities, area scope and canonical URLs in findings. Use only URLs returned by the tools for CostInflation hyperlinks; use links.methodology_url for general methodology and the returned category URL for category-specific methods. Never invent a path or append a guessed path or fragment. Ask for clarification when categories or units are ambiguous. Do not infer missing history, causality or national coverage. Keep daily history statistics distinct from underlying product distributions. Package matches cover consecutive days with unchanged identity/title/brand; describe the matched cohort and avoid inferring shrinkflation or intent. Never sum marginal percentiles into basket percentiles.