MPEP and TMEP API: Patent and Trademark Examining Procedure You Can Query

Short answer: the MPEP and the TMEP are the USPTO's own instruction books for its examiners, and you can query both over an API by manual section number instead of scraping uspto.gov. The Manual of Patent Examining Procedure runs 2,024 sections across 29 chapters. The Trademark Manual of Examining Procedure runs 2,109 sections across 19 chapters. Neither is law. Both are the single best predictor of what an examiner will actually do with your application, which is why practitioners cite them harder than they cite the statute. This post covers what they are, how they cite, how to filter them, and how to walk from an office action to the statute and rule underneath it in one chain.

TL;DR

• The MPEP is 2,024 sections in 29 chapters; the TMEP is 2,109 sections in 19 chapters. Both are USPTO guidance, not binding law, and both are the operating manual an examiner follows.
• They cite by manual section, not by CFR part. "MPEP 2106" and "TMEP 1207.01" are addresses inside a manual, so a CFR-shaped parser will not find them.
• US Copyright Office Circulars are the third piece of the IP guidance set: 58 circulars indexed as 704 retrieval passages, written for applicants rather than examiners.
• All three sit behind corpusType=AGENCY_GUIDANCE and split with source=mpep, source=tmep, and source=copyright_circular. Refresh is weekly.
• The useful move is the chain, not the lookup. An office action cites a manual section; the manual section rests on a statute in 35 U.S.C. or 15 U.S.C. and a rule in 37 C.F.R. You want all three in one call sequence.
• Versioning here is the amendment record, not a date parameter. The corpus holds one current text per citation, plus amendment history. If you need the MPEP as it read in 2019, this is not that.

The IP entry in our federal agency guidance series.

For related coverage, see Federal Agency Guidance API: 34 Sub-Regulatory Sources in One Endpoint, USCIS Policy Manual API: 456 Chapters of Immigration Guidance, Queryable by Volume for the same manual-shaped problem in immigration, and Export Controls API: BIS Advisory Opinions and DDTC Commodity Jurisdiction Determinations.

Disclosure: we build Vaquill AI. Every count below is our own, measured against the live index on 2026-08-30, and you can check the totals yourself against the free GET /us/statutes/coverage endpoint.

What the MPEP actually is

The MPEP is the book a patent examiner is trained on and reviewed against. It tells the examiner how to read a claim, when to issue a restriction requirement, what a proper 35 U.S.C. 112 rejection looks like, and how to write the rejection up.

It is published by the USPTO, revised on the Office's own schedule, and organized in a 29-chapter spine keyed to hundreds-blocks. Chapter 700 covers examination of applications. The 2100 block covers patentability, which is where most prosecution arguments live.

The MPEP is not law. It is the Office's reading of 35 U.S.C. and 37 C.F.R. A court is free to disagree with it, and courts have. An examiner is not.

That gap is the whole reason it matters to a docketing or prosecution tool. If you want to predict the next office action, the statute tells you what is possible and the MPEP tells you what is likely.

MPEP 2106 is the canonical example. It is the subject matter eligibility section, and it is where the Office translates a line of Supreme Court cases into a decision procedure an examiner can run. A response that argues eligibility without addressing 2106 on its own terms is arguing past the person reading it.

What the TMEP actually is

The TMEP is the same idea on the trademark side: 2,109 sections across 19 chapters, written for examining attorneys at the USPTO.

It covers what makes a specimen acceptable, how to handle a disclaimer, when to require an amended identification of goods, and the refusal grounds. TMEP 1207.01 is the likelihood-of-confusion section, which is the ground behind a large share of substantive trademark refusals.

Same legal weight as the MPEP: internal guidance, binding on examining attorneys, persuasive at best in front of a tribunal. Same practical weight: it is what the letter you just received was written from.

The third piece: US Copyright Office Circulars

Copyright rounds out the IP guidance set, and it works differently. The Copyright Office does not run an examining corps the way the USPTO does, so its public guidance takes the form of Circulars: short, numbered, plain-language explainers on one registration or practice topic each.

A Circular is not a rule and not a manual section. It is the Office telling applicants how it handles a category of work, what a deposit needs to contain, and which application path fits. Circular 1, "Copyright Basics", is the front door of the set.

The corpus holds 58 circulars, indexed as 704 retrieval passages. That passage count is higher than the document count because a long circular splits into several passages for retrieval, which is how the whole index works.

One boundary worth naming. Circulars are the short public explainers, and the Compendium of U.S. Copyright Office Practices is the long internal manual. They are separate document sets, and the Compendium is not part of this 58-circular collection. If your workflow depends on Compendium sections specifically, check coverage before you build against it.

The three manuals side by side

All three share corpusType=AGENCY_GUIDANCE and refresh weekly. The corpus types reference lists every token the API accepts.

The citation form is the integration problem

This is where most IP tooling breaks, so it is worth being precise.

A CFR citation is title-part-section: 37 C.F.R. 1.111. A US Code citation is title-section: 35 U.S.C. 101. Both have a shape a parser can key on, and both map cleanly onto a titleNumber plus part filter.

A manual citation has none of that. "MPEP 2106" is a bare number that is meaningful only inside the MPEP. It is not a CFR part, it is not a title, and it does not appear in the US Code.

Manual sections also nest by suffix. The MPEP goes to forms like 706.02(a), and the TMEP goes deeper still, with lettered and roman-numeral subsections hanging off a section like 1207.01. A regex written for \d+\s+U\.S\.C\.\s+\d+ will find none of it.

The practical consequence: if your citation extractor was built for statutes and regulations, it is silently dropping every manual cite in every office action you ingest. That is usually the highest-value citation in the document, and the Bluebook forms a parser can and cannot handle is the longer argument.

Read that diagram bottom-up and you have the response strategy. The examiner reasons down from the manual. You answer at the manual, then reach for the statute when the manual is being applied past its own terms.

Corpus mechanics

The whole federal agency guidance corpus is 21,906 sections across 34 named sources, and the three IP manuals are three of those sources. Everything is reached through one search endpoint.

Base URL is https://api.vaquill.ai/api/v1. Auth is a bearer key. The search call costs 4 credits and takes corpusType plus source as the two filters that matter here. You can run it in the API playground first.

curl -X POST https://api.vaquill.ai/api/v1/us/statutes/search \
  -H "Authorization: Bearer vq_key_..." \
  -H "Content-Type: application/json" \
  -d '{
    "query": "abstract idea judicial exception eligibility analysis",
    "corpusType": "AGENCY_GUIDANCE",
    "source": "mpep",
    "matchType": "any",
    "limit": 10,
    "excerptChars": 600
  }'

Three behaviors worth knowing before you wire this into a job:

• Unknown filter values return 422, not an empty list. If you typo source: "mpeps", you get an error, not a confident zero, and the error reference has the full catalogue. That difference has saved more than one nightly job from reporting a clean run over nothing.
• The source error message lists every valid code. So the fastest way to discover the filter vocabulary is to send a wrong one on purpose.
• Failed calls are not charged. A 422 costs nothing.

Paging is cut from a single ranking, so results do not repeat or vanish between pages, and page 4 costs the same as page 1. If you are sweeping all 2,024 MPEP sections into a local mirror, that matters, and the browse hierarchy recipe is the walk-the-chapters pattern.

Worked example: an office action to its authority chain

Here is the sequence a prosecution tool should run when an office action lands citing MPEP 2106.

Step 1: find the manual section. Search the guidance corpus scoped to the MPEP.

{
  "query": "2106 subject matter eligibility",
  "corpusType": "AGENCY_GUIDANCE",
  "source": "mpep",
  "matchType": "any",
  "limit": 5
}

Each hit comes back with an actId, the citation, an excerpt, and the official uspto.gov URL for the publisher's own copy. Read the actId off the response and store it. Do not build one from the citation string. Ids are stable and not guessable, and hand-built ones usually 404.

Step 2: pull the full text. With the id in hand:

GET /us/statutes/section/{actId}
GET /us/statutes/section/{actId}/body

The first gives you the section with its metadata. The second gives you the body text on its own, which is what you want if you are feeding a model rather than rendering a page.

Step 3: get the statute underneath. MPEP 2106 is the Office's procedure for 35 U.S.C. 101, so pull the statute itself out of the USC corpus.

{
  "query": "inventions patentable process machine manufacture composition of matter",
  "corpusType": "USC",
  "titleNumber": "35",
  "limit": 5
}

The USC corpus is the 2024 edition, current through 2025-01-06, at 60,170 sections. It is checked weekly for a new edition. That currency date is published rather than implied, which is the point: you can tell a reviewing attorney exactly what edition your tool answered from.

Step 4: get the rule underneath. Procedure for replying lives in 37 C.F.R. Part 1, inside the CFR corpus.

{
  "query": "reply to a non-final office action requirements",
  "corpusType": "CFR",
  "titleNumber": "37",
  "part": "1",
  "limit": 5
}

The CFR corpus is 219,114 sections and refreshes daily, which is the fastest cadence in the federal set. For a docketing product that is the right property: rule changes that move a deadline should not wait a week.

Step 5: hand the model all three. You now have the manual section, the statute, and the rule, each with its own source URL. That is a citation-checkable answer instead of a plausible one.

The chain is the product. Any tool can retrieve MPEP 2106. Retrieving 2106 alongside the statute it implements and the rule that governs the reply is what makes a drafted response defensible.

Monitoring: the part that actually changes

The MPEP and TMEP move on the Office's revision schedule, not on notice-and-comment. A revision can change examiner practice without a single word of 37 C.F.R. changing.

This corpus refreshes weekly, and changes captured on each refresh become change events. GET /boards carries no charge and hands back every watchable source with three fields attached: cadence, retrievalStatus, and lastRetrievedAt. That last one tells you when a source was actually pulled, not when it was supposed to be.

From there you can create a watch, list changes, and pull a per-section diff. For an IP practice, the high-value watches are narrow: the eligibility block of the MPEP, the refusal-grounds block of the TMEP, and 37 C.F.R. Parts 1 and 2.

There is a related read on how change detection and alerts work if you want the mechanics, and a longer piece on diffs and false positives.

Who this is wrong for

Say the honest part. This corpus is examining procedure and public guidance. It is not the adjudicative record.

• It is not PTAB or TTAB decisions. Inter partes review outcomes and opposition proceedings are not here.
• It is not Federal Circuit case law. MPEP 2106 summarizes a line of cases; it does not contain them.
• It is not file wrappers. Nothing about a specific application's prosecution history is in a corpus of manuals.
• It is not a substitute for reading the revision notes when the Office issues a new MPEP revision.

If your product needs PTAB final written decisions, this is not the dependency you are looking for. If your product needs to know what an examiner is required to do before writing a 101 rejection, it is.

FAQ

Is the MPEP legally binding?

No. The MPEP is the USPTO's internal guidance to its own examiners, and it does not have the force of law the way a statute in 35 U.S.C. or a rule in 37 C.F.R. does. Courts are free to reject the Office's reading and sometimes do. Examiners are not free to ignore it, which is why it drives outcomes at the prosecution stage.

What is the difference between the MPEP and the TMEP?

Both are USPTO examining manuals, split by right. The MPEP covers patent examination in 2,024 sections across 29 chapters. The TMEP covers trademark examination in 2,109 sections across 19 chapters, including refusal grounds like likelihood of confusion at TMEP 1207.01.

How do you cite an MPEP section?

By manual section number, with no title or part: "MPEP 2106". Subsections hang off that number with letters and roman numerals, so a cite can look like MPEP 706.02(a). This is why a citation parser built for the CFR or the US Code misses manual cites entirely.

What is a US Copyright Office Circular?

A Circular is a short, numbered explainer the Copyright Office publishes on one registration or practice topic, written for applicants rather than for internal staff. Circular 1 covers copyright basics. The corpus holds 58 circulars, indexed as 704 retrieval passages.

Can I get the MPEP as it read on a past date?

Versioning here runs off the amendment record rather than a date parameter. Each manual section is stored at its current revision and only that one, and no as_of parameter is offered. Alongside it you get amendment history, a yearFrom and yearTo currency filter, and diffs on watched sources, so a prior MPEP revision has to come from the USPTO's own archived revisions.

How often does this data refresh?

Federal agency guidance, which includes the MPEP, the TMEP, and Copyright Office Circulars, refreshes weekly. The CFR underneath them refreshes daily. The US Code corpus is the 2024 edition current through 2025-01-06 and is checked weekly for a new edition.

Do I need separate API calls for the manual, the statute, and the regulation?

Yes, and that is deliberate. Each call is one search against one corpus, scoped by corpusType, so the chain from MPEP 2106 to 35 U.S.C. 101 to 37 C.F.R. Part 1 is three scoped calls rather than one fuzzy one. Each result carries its own official source URL, so the resulting answer is checkable line by line.

Are PTAB and TTAB decisions included?

No. This corpus is examining procedure and agency guidance, not adjudicative decisions. PTAB inter partes review outcomes, TTAB opposition decisions, and Federal Circuit opinions are all outside it.

Why does the passage count differ from the section count?

A section is one citable item with a stable actId. Long documents split into several retrieval passages so semantic search can hit the right part of a long text. That is why 58 Copyright Circulars index as 704 passages.

What happens if I send an invalid source value?

You get a 422 with an error message listing every valid source code, and the call is not charged. Unknown filter values are rejected rather than silently matching nothing, so a typo in a nightly job fails loudly instead of reporting an empty but successful run.

Where to start

If you are building IP tooling, the first call to make is the free coverage endpoint, then one scoped search against source=mpep with a query drawn from a real office action you already have. You will know inside ten minutes whether the retrieval quality clears your bar.

Priyansh Khodiyar

Co-Founder & CTO

Priyansh leads engineering and AI at Vaquill, from the matter workbench to drafting, document comparison, document matrix, and citation-verified research.