Data Pitfalls for Claude

An MCP connector that gives Claude the 78 rules from Avoiding Data Pitfalls by Ben Jones (Wiley, 2020), so it can check your SQL, charts, dashboards, written analysis and reports for the mistakes that mislead.

How it works

The connector serves the pitfall taxonomy and nothing else. Claude asks it for the checklist that fits the work in front of it, checks each item against your work itself, and reports what it finds with the rule ids and fixes. Your work stays in your conversation with Claude: the tools never receive your data, charts or queries. The connector makes no AI calls of its own, stores nothing, and needs no account or API key. Every tool is read-only.

Connect it

claude.ai and Claude Desktop: open Settings → Connectors → Add custom connector, name it Data Pitfalls, and paste this URL. No sign-in is needed.

https://www.avoidingdatapitfalls.com/api/mcp

Claude Code: either use the same URL,

claude mcp add --transport http datapitfalls https://www.avoidingdatapitfalls.com/api/mcp

or run the server locally from npm:

claude mcp add datapitfalls -- npx -y datapitfalls mcp

Use it

Ask in your own words. For example:

“Check this chart for data pitfalls.” (with an image attached)
“Review this SQL query for data pitfalls before I share the numbers.”
“Build me a dashboard of monthly signups, then review it for data pitfalls and fix what you find.”
“What does the truncated y-axis pitfall look like, and how do I avoid it?”

Claude reports each finding as active (evident in the work) or latent (depends on data it can’t see), rates how much it matters, quotes the evidence and gives the fix. A clean review means none of the cataloged pitfalls were found, not that the work is correct.

There is also a review_for_pitfalls prompt, and each rule is available as a resource at datapitfalls://rules/{id}.

Tools

`get_checklist`	The rules that apply to a kind of work (sql, code, chart, image, prose, text, dashboard or document) as a checklist, each with what to check for and how to fix it. Optionally narrowed to some domains.
`get_pitfall`	One rule in full by id: what it is, how to detect it, an example of the pitfall, an example done well, the fix and the book reference.
`list_pitfalls`	All 78 rules, optionally filtered by domain or severity.
`search_pitfalls`	Keyword search over the rules, e.g. "dual axis" or "missing values".

Privacy and support

See the privacy policy. For questions and problems, open an issue on GitHub. The connector is open source: bjonesdataliteracy/datapitfalls.