Patent Bear insideClaude,ChatGPT and Codex,and Gemini.

Search, Research, Draft, and Library, as tool calls in your agent. Listed in the Claude connector directory. Connect now

Patent Bear is a patent research connector for AI agents. Connect it once and Claude, ChatGPT, or Gemini can pull real patent records, prosecution documents, citations, families, court decisions, and examiner office actions, quote them verbatim with column and line cites, run patent analysis, and save the work product back to your Patent Bear Library.

It exposes 47 tools at https://www.patentbear.com/mcp. It is built for patent attorneys, patent agents, paralegals, and IP analysts, and it is listed in the Claude and ChatGPT connector directories.

The record

Read the patent itself

Your agent reads the actual record rather than recalling it. Text comes back verbatim, and passages in issued grants carry the column and line cites you can put in a filing.

get_patent_record

The full text of a US or European (EP) grant or publication by number: title, claims, and description, plus bibliographic data, assignee, inventors, examiner and art unit, CPC, and the continuity chain.

Also returns term data: patent term adjustment with its A, B, and C breakdown, an expiration estimate, terminal-disclaimer status, and double-patenting rejections given or received. Every description paragraph in a grant carries its column and line range, so an agent can read the whole specification and cite as it goes.

find_in_patent

Locate text inside one patent and get the verbatim passages with column and line cites, in the form 4:12-19. Quote ready for invalidity contentions, claim charts, and office-action responses.

Search by terms, or pass an exact sentence you already have to get its precise lines, or pass up to 50 sentences at once and get every cite in one call. Every passage carries the citation string to use, what the range actually bounds (sentence, paragraph, or snippet), and a flag when a cite is approximate and needs confirmation against the issued document.

get_term_definitions

The patent dictionary: 94.8 million explicit definitions mined from US patents and published applications, 1976 to today. The sentences where a specification actually defines a term, plus every abbreviation a patent spells out.

Ask how one patent defines its terms, or how the whole corpus has defined a term, filtered by assignee, date range, and grant or publication. Useful when drafting a definitions section or testing a proposed construction against how others have used the word.

search_patents

Search patents, published applications, and technical literature by number, by keyword, or by date window. Hits carry title and abstract as verbatim text, with matching snippets from the claims and specification.

US grants and published applications by default. Ask for European patents and it searches the EP corpus instead, 8.1 million EP grants and publications in full text from 1980, or both offices together.

search_patents_bulk

The same search returned as a long compact list, up to 1,000 results a call, for landscapes, portfolio sweeps and counting.

Every answer states how many records matched in total and how to get the next block, with counts by owner, classification and year across the whole set. Results can be ordered by relevance, newest or oldest, across US, European, or both. One request is counted per 25 results returned.

find_similar_patents

The patents whose claims read closest to a patent's independent claims, or to any text you supply: a claim, a single element, or a sentence describing the idea.

Nearest-neighbour search over 47.6 million claim embeddings. It finds neighbours to read, not prior art: there is no date cutoff and no classification filter, so for art against a claim with a priority date, run a Prior Art analysis instead.

get_assignments

USPTO recorded assignments: for a patent, publication or application, the chain of recorded conveyances newest first with the latest recorded assignee; for a company or person, every record naming it as assignee or assignor.

Recorded, not adjudicated: the tool says latest recorded assignee, never owner, and marks which records transfer title and which are security interests, licenses or releases.

search_applications

US applications by what the Office's own record says: applicant, inventor, examiner, art unit, classification, title words, filing window and status.

Status groups: pending, granted, abandoned, allowed, rejected, appeal. Reads the application record rather than the full-text index, so it answers questions about prosecution and status; technology questions go to search.

get_orange_book

FDA Orange Book patent listings by patent, drug (trade name or ingredient), NDA holder or application number: use codes, the expiry the holder submitted, pediatric extension and delisting flags.

get_article

The full text of one open-access journal article from PubMed Central, by PMCID, DOI or PMID, with its citation fields, license and reference count.

get_figures

The drawings of a US or EP grant or publication: every sheet, the record's own description of each figure, and the sheets themselves as images your AI agent can look at.

Ask for up to four sheets at a time. Older US grants have no drawing files, so their sheets are extracted from the issued PDF on first request; the tool says when that is under way and to ask again in a few minutes.

Prosecution

Trace what happened at the Office

The file wrapper, the examiner's own words, and the art that was placed on the record, for prosecution history, estoppel context, and double-patenting questions.

get_file_history

The prosecution timeline for a US grant or publication: the file wrapper grouped into dated events, each marked applicant or USPTO and flagged when it is a rejection or an allowance decision.

Returns the post-grant picture in one place: issue date, PTA days, expiration estimate, terminal-disclaimer status, double-patenting rejections, and any reexamination or AIA-trial certificates. Reexamination papers live in their own wrappers and are read by passing the control number. The response says when the wrapper was last read, so an agent can tell the user that date and offer to refresh rather than assume the timeline is current.

get_file_history_document

The text of one file-wrapper document: an office action, an amendment, the applicant's remarks, a claims listing, an examiner's reference list.

Use it before characterizing prosecution. Which statute a rejection rests on and over what art, what the applicant amended and what it argued to distinguish that art, and what the examiner said in allowing. Scanned documents are read by OCR on first request.

search_office_actions

Search 13 million examiner office actions, non-final and final rejections, from July 2007 to the current week, by the examiner's own words and by the USPTO's rejection tags.

Filter by rejection type (101, 102, 103, 112, double patenting), tech center, art unit, examiner, document code, year, application, or patent. Answers questions like how a given examiner has applied section 101 to claims of this kind. Examiner side only: pair it with the file wrapper for the applicant's response.

get_citations

The front-page References Cited: the art placed on the record during prosecution, both the applicant's information disclosure statements and the examiner's own citations.

Forward citations (later patents citing this one) and consolidated family citations are also available on paid plans. Note the distinction the tool enforces: a reference disclosed to the examiner may never be mentioned in the specification body.

get_family

The US continuation and divisional tree with application numbers, or the worldwide family from USPTO Global Dossier: every member application by office, its publications, filing dates, and the priority claims linking them.

Litigation

Check whether it has been fought over

District court dockets, PTAB proceedings, post-grant certificates, and the decisions themselves, read page by page from the court's own PDF.

get_legal_activity

Whether a US patent has been litigated or challenged, in one call: district court cases from 1963 dockets to the present, outcome buckets for terminated cases, and PTAB proceedings with status and petitioner.

Also returns certificates from Patent Bear's certificate corpus: reexamination and AIA-trial outcomes showing which claims were cancelled, confirmed, amended, or added, and when. That matters before any validity work, because cancelled claims should not be charted.

search_litigation

Search about 58,000 US patent court decisions and orders from the district courts, the Federal Circuit, and the PTAB era, 1965 onward, parsed from govinfo.

Search by party, judge, court, patent in suit, law firm, cited authority, case number, doctrine, disposition, outcome, year range, or words in the opinion text. Results carry the reporter citation when the decision was published, a one-line holding, and links to both the Patent Bear reader and the court's PDF.

get_opinion

One decision in full, page by page as parsed from the court's PDF, with the reporter citation, footnotes, parties, judge, patents in suit, and disposition.

Before a pinpoint cite goes into a filing, the agent can pass the quote and get back the page the words are actually on, correcting an off-by-one and warning when the stored text dropped a hyphen so the quote prints as the PDF has it. The original PDF can be returned alongside the text for confirmation.

Analysis

Run a Patent Bear analysis

Analyses run on Patent Bear's servers, exactly as if started from the site, and land in the account's Library. Analysis runs consume account credits; sequence, chemistry, and citation searches are included with Pro and Team and cost one credit on Plus.

run_research

Start an analysis. Prior Art is invalidity prior-art search with element-by-element claim mapping. Disclosure is section 112 written-description and enablement support. Charting maps named references against the claims. Citations analyses a granted family's forward and backward citations. Sequence and Chemistry search by homology or structure, from a patent or a pasted sequence or SMILES. Clearance screens a product description against active claims. Invention searches prior art from a plain-text disclosure.

Every run is deliberately two calls, like the claim setup on the site. The first returns the intake and starts nothing: for Prior Art, the operative claim set after any reexamination or AIA-trial certificate, each claim's status and text, what a selection costs, and the constructions already saved on the patent; for the other analyses, what will run and what it costs. The agent walks the user through it and never picks the claims for them; only the second call, after the user confirms, spends a credit. Follow-up gap and broad searches re-run retrieval aimed at the limitations that went unmapped, with the same confirmation.

get_research_run

Check a run and read its results once it is done: per claim, the best reference with its coverage and the verbatim evidence quote for each element, the gap-filler, the proposed combination and its remaining gaps, other charted candidates, and the search funnel.

Results are built to be read honestly. Each element carries its support level, strong, thin, generic context, or not found, and names the clause the citations do not reach. Coverage is triage rather than a conclusion, and a run that leaves elements unmapped is a search result, never a validity opinion.

list_research_runs

The account's analysis runs of every kind, including runs started in the browser, newest first, each with status and a link to the results.

get_claim_chart

One claim chart in full: every charted claim broken into elements, and for each element the evidence written under every reference column, with quotes, column and line cites, and figure callouts.

list_claim_charts

The claim charts in the Library with target patent, matter, origin, and last edit. Charts open in the Library, editable and exportable to Word.

Draft

Build filing-ready documents from the record

Each draft is one call: an input goes in, a Word document comes back and lands in the Library, where it can be downloaded again at no charge. Every field is drawn from the record or from what you pasted; the model never writes a field value.

draft_definitions

A Definitions section for a new specification from the client's own earlier filings. Give it the terms, or a patent whose independent claims supply them, and the client. Each definition is the client's most recent prior wording, verbatim, with a Word comment naming the source patent.

Where the client has several definitions the alternatives are listed in the comment; where it has none the paragraph is a placeholder with reference definitions from other assignees; a court construction saved in your tracker is flagged. One model call per term selects which prior definition fits the sense and never rewrites it. One credit per run.

draft_ids

An Information Disclosure Statement from any pasted list: SB/08a rows for US patents, US publications and foreign documents, SB/08b rows for non-patent literature, plus a CSV for Patent Center's auto-load form.

US, EP and CN/JP/TW rows are filled from Patent Bear's records: number, kind code, date, first inventor or applicant. Other offices pass through with the cells to confirm, non-patent citations pass through as typed, duplicates collapse, and every line that did not resolve is listed verbatim so nothing is silently dropped. Included with the plan.

draft_double_patenting_chart

A patent's claims charted against its commonly owned US family members: the family table with terminal-disclaimer and restriction evidence, examiner double-patenting rejections quoted from the office-action record, two-column claim charts with shared phrases marked, and the ordering facts.

The chart maps claim language and makes no obviousness-type determination. Included with the plan.

get_draft

Re-open a draft: status, the Word download link, the draft page, and the data behind the document.

list_drafts

The account's drafts, newest first, filterable by kind and patent.

Work product

Keep what the agent found

Everything can be written back to your Library, where it renders exactly as if you had done it by hand. Notes an agent writes are bylined with the app that wrote them and never overwrite your own.

save_highlights

Highlight passages in the Patent Bear reader, exactly as selecting a passage and clicking Highlight does, optionally with a note. Also saves claim terms tracked across the patent, and notes about the patent as a whole.

Notes are versioned into named sets, so a set name like Invalidity chart v2 collects one piece of work across every document it touches. The tool reports back what was actually saved and flags a partial match, so an agent cannot claim a passage was highlighted when only part of it matched.

save_opinion_highlights

The opinion twin of save_highlights: highlight passages in a court decision by page, each optionally with a note, saved as your own annotations on the decision's reader page.

save_claim_construction

Save a term and its construction to the Constructions tracker, scoped to a patent and optionally a matter, with the claims it applies to, the intrinsic or extrinsic basis, the construction type, and its status: proposed, agreed, court, or disputed.

save_definitions

Save a definitions document to a matter. Reviewers can select terms, compare source wording, edit, and download Word with source comments and PatentBear links.

list_documents

Find definitions and claim charts across your accessible matters. The Library holds the list; each document opens in its own editor tab.

get_document

Read the latest saved document before revising it, including selected terms, source quotations and its version, so an agent can preserve the user’s changes.

write_note

Write a long-form note into a matter's notebook in markdown: a case summary, a prosecution and litigation history, a research memo, a call note. Highlights live on the patent, notes live on the matter.

save_definitions

Save a drafted definitions section into a matter's notebook as a structured note and get a Word file where every definition carries a comment with its provenance: the verbatim source sentences, the patents they came from, how many of the applicant's own definitions took that form, and any divergence flag.

Use it for a section the assistant drafted itself; draft_definitions builds one from the dictionary in a single run. Each entry names the term, the proposed definition, and its sources from the patent dictionary, or is marked original when no prior definition existed. Paid plans only.

list_notes

The notes in a matter's notebook, with titles, ids, and links, so an agent revises the right note instead of starting a second one on the same subject.

list_annotations

Your highlights, claim terms, and constructions for one patent, or the highlights on one court decision, each with a link that opens the reader on that passage.

Notes come grouped by named set, showing who created each, you or an agent by app name. A set name on its own lists every patent and decision it touches plus one link that opens them all, which is the link to hand back for a brief.

save_patent

Save a patent to the Library, optionally filing it into a matter and attaching a note.

save_opinion

Save a court decision to the Library under Opinions with its caption cached. Every decision relied on for a brief, chart, or memo should be saved and its quoted passages highlighted.

save_patent_analysis

Write a structured Overview or Claim Summary onto a patent's Library page, attributed to the connector. The agent authors it from the patent's real record, with verbatim grounding rather than memory.

get_patent_analysis

Read the stored Overview and Claim Summary analyses for a patent, whoever wrote them, with model provenance and timestamps.

list_saved_patents

The account's saved patents, newest first, optionally limited to one matter, each with its note and a link.

create_matter

Create a client-matter folder by name. Idempotent: an existing matter of the same name is returned rather than duplicated.

list_matters

The account's matters with ids, names, and item counts, so a matter the user names resolves to the right folder.

help

A guide to everything the connector does, tool by tool, with example prompts a user can say. An agent calls it when it is unsure which tool fits, so it can answer what Patent Bear can do without guessing.

Getting started

How to connect

Add https://www.patentbear.com/mcp as a connector in Claude, ChatGPT, or Gemini, then sign in to your Patent Bear account when prompted. Patent Bear is listed in the Claude and ChatGPT connector directories, so it can also be added from there.

The Connect page walks through six worked examples, and the guides cover the same work done on the site.

Getting started

Plans and limits

The free plan includes 20 connector requests per month. Reading the record, searching litigation, searching office actions, term definitions, and similar-patent search are available on every plan. Saving to the Library, claim charts, and forward and family citations are paid features, and a free-plan request for one returns an upgrade link rather than failing silently. Analysis runs consume account credits; sequence, chemistry, and citation searches are included with Pro. Plan prices are shown once you sign in.

What to expect, in the same words as the directory listing: source coverage and citation precision vary, the tools flag unavailable and approximate evidence, and work product requires attorney review. Retrieval results are interpreted by your chosen AI service. Requests and submitted passages reach Patent Bear; saved work is stored in your account, and matter notes are visible to authorized members.

See pricing, privacy, and confidentiality for the full terms.