DingDuff: A Claude MCP connector for legal research

I'll pull the product page and HN thread so the paragraphs stay factual and specific.Fetching the benchmark page for concrete comparison details.DingDuff is a free Claude MCP connector for legal research, built by practicing lawyers Kyle Dingman and Stephanie Duff-OBryan for their own work and now open to licensed attorneys. It links an existing Claude account to primary legal sources—millions of opinions (via CourtListener), federal and state statutes and rules, and court filings—plus citation support from The Indigo Book, Second Edition (2023). Setup is a short Claude connector-and-plugin flow; the product is funded by a tip jar rather than a traditional SaaS subscription. It appeared on Hacker News with a link to dingduff.com and early traction (1 point, 0 comments at listing time).

On the technical side, DingDuff is deliberately thin: an MCP connector and Claude plugin that expose tools and skills so Claude does the research, rather than a closed legal chatbot that reimplements research UX. Depth depends on how Claude can handle sources. The preferred path is fetch_opinion_file: Claude downloads full opinions, saves them as files (ideally a local folder in Claude Cowork, or a virtual machine on web/mobile with code execution and network egress on), and reads the full text. A backup tier, opinion_extract, pulls key passages and needs a bring-your-own Anthropic API key (roughly $5–15/month to Anthropic for heavy use). A weaker fallback, opinion_view, keeps case text inside the chat. The founders report that Claude with DingDuff does not invent fake cases or statutes in their experience, with the lawyer as the verification layer—and access is limited to lawyers under their terms.

For engineers and builders, the interesting claim is architectural: a general frontier model plus direct, structured access to primary sources can beat or match purpose-built legal AI products that wrap retrieval and guardrails in a proprietary product. That is an MCP-and-tooling story as much as a legal one—connector design, file I/O, network egress, and skill packaging matter more than a polished research UI. The same pattern generalizes to any domain where the hard part is trustworthy source access and citation checking, not model personality: compliance, medical literature, finance filings, or internal policy corpora.

Competitively, DingDuff sits against Westlaw CoCounsel and Lexis Protégé, and against other frontier models used without a legal connector. On one shared assignment—a multi-issue Delaware/Texas LLC veil-piercing and TUFTA memo—the team hand-graded every citation. Claude with DingDuff (reported as Fable at high effort) scored 11/11 on substantive sub-issues and 131/131 evaluated cites correct; ChatGPT scored 9/11 and 50/51 cites; Westlaw CoCounsel 9/11 and 87/91; Lexis Protégé 4/11 and about 76% of evaluated cites correct. Only the Claude + DingDuff run flagged Texas’s statutory bar on reverse veil-piercing for LLCs under BOC § 101.112(d). The comparison is one prompt, graded by the product’s own lawyer-founders, so treat it as a directional stress test, not an industry-wide ranking—but it is unusually transparent, with memos and cite panels published for inspection.

Practically: if you ship MCP tools for professional work, optimize for full-source retrieval and local or VM file access, not chat-only summaries; document account settings (egress, code execution, folder attach) because they change research quality; and keep a paid fallback path when the best tool is unavailable. Watch whether independent lawyers reproduce the Westlaw/Lexis comparison outside the founders’ grading, how CourtListener coverage gaps show up in state practice, and whether Anthropic’s connector directory and Cowork file model become the default distribution path for vertical research tools of this type.