TaxCite v0.1.0 — first public release

The first public release of TaxCite: a command-line tool and Model Context Protocol server that retrieves U.S. federal tax law and verifies the citations and quotations in AI-drafted tax writing against the official government sources.

Website and documentation →

What it does

Given a draft, TaxCite extracts every citation, resolves each against the source it belongs to, and checks any quoted language attached to it. It separates failures that look identical on the page but need different corrections:

Finding	Meaning
verified	The provision exists and the quotation is verbatim
pinpoint_not_found	The section is real; the subdivision cited is not
quote_close	Altered from the source — with a word-level diff
quote_misattributed	Genuine language, wrong provision
quote_not_found	Appears nowhere in the source
unverifiable	Not checked — never confused with "wrong"

Output is Markdown for a reader, JSON for a pipeline, or GitHub annotations for CI.

Sources

Internal Revenue Code (USLM XML, Office of the Law Revision Counsel) · Treasury Regulations (eCFR) · IRS guidance from the Internal Revenue Bulletin, with a citator built from the supersede and obsolete relations the Bulletin itself states · federal case law through CourtListener.

All are works of the United States government and in the public domain. No proprietary research service is scraped, queried, or depended upon.

Beyond existence

• Authority classification under Treas. Reg. § 1.6662-4(d)(3)(iii)
• Effective-date checking from the formulas in statutory notes (--tax-year)
• Temporary regulation sunset under I.R.C. § 7805(e)
• Historical release points, to check a draft against the law as it stood (--as-of)
• A diligence record identifying a document by hash rather than content
• An enforced --offline mode that seals the process against every outbound request

What it deliberately will not do

It does not judge whether a legal conclusion is correct, and it does not check whether a decision is still good law — treatment signals are editorial products no free source publishes, and the available heuristics are wrong often enough to be worse than silence. Where a source's coverage is incomplete, a miss is reported as unverifiable rather than as an error: failing to find something you never looked for is not evidence.

⚠️ Experimental research software

Not tax advice, not legal advice, and not a substitute for professional judgment or a commercial citation service. Provided "AS IS" without warranty or condition of any kind, and without any guarantee of accuracy, completeness or currency. No part of this project has been reviewed by a licensed tax practitioner. You are responsible for independently verifying every authority you rely on.

---

Apache License 2.0 · Copyright © 2026 Muhammad Rakibul Islam · rakib.islam@rutgers.edu