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 Connect · in the Claude and ChatGPT directories

Patent Bear insideClaude, ChatGPT,
and Gemini

Structured patent data, for your research and for your agent. Quotes with the column and line, the file wrapper, the court decisions.

Free to start: 20 connector requests a month.

01 / In action

Real prompts, real results.

Real runs in Claude. Open a result.

Brief-ready quotes.1 of 6
ClaudePatent Bear connected
Create a table that quotes every specification passage of US 7,844,915 supporting the claim terms “event object”, “scroll or gesture call”, and “rubberbanding”, each cited as (’915 Patent, col:line).
✓find_in_patentUS 7,844,915 · 3 terms · col:line

Every supporting passage, quoted and cited, as a Word table.

915-spec-support-passages.docxDownload

02 / From answer to record

The answer, and the record behind it.

A real exchange in Claude on the left. The same passages, lit in the reader, on the right.

ClaudePatent Bear connected
Quote every passage of US 7,844,915 supporting “scroll or gesture call”, “rubberbanding”, and “event object”, cited col:line. Save each as a highlight.
Patent Bear reader with the scroll or gesture call passage of US 7,844,915 highlighted at 5:12-15Patent Bear reader with the rubberbanding passage of US 7,844,915 highlighted at 2:15-18Patent Bear reader with the event object passage of US 7,844,915 highlighted at 12:30-32
’915 at 5:12-15See the demo on desktop

26 highlights saved, each with the cite as its note.

Line numbers are read from the printed gutter. Confirm against the issued PDF before filing.

Each cite opens the reader on that passage, highlighted, with the note. Confirm it yourself before it goes in a brief.

03 / Under the hood

What your agent sees.

One patent. Every piece, structured and cited.

  1. 01BibliographyNumber, dates, assignee, status, family.US 7,844,915 B2
  2. 02ClaimsEach claim, each element, numbered.claim 1, element 1c
  3. 03SpecificationFull text, sentence by sentence.col. 13, ll. 37–40
  4. 04FiguresEvery sheet, with the reference numerals.FIG. 3, ref. 310
  5. 05Column and lineEvery quote comes with its cite.verified, not approximate
  6. 06File historyOffice actions, responses, allowance.OA 2009-03-12, resp. 2009-06-11
  7. 07Citations and familyForward, backward, and the continuations.12 cited · 340 forward · 4 in family
  8. 08Decisions and definitionsCourt rulings and construed terms, to the page.Fed. Cir. 2015, slip op. at 9

04 / Our data

Patent data, focused and reimagined.

Nine purpose-built data sets, processed and derived from the record for the questions agents get asked. Only on Patent Bear.

Patents · data set 01 / 09

US grants and publications

14.5M+ US grants and published applications, full text: every claim, the description with a column:line cite on every paragraph, and every figure sheet with its reference numerals.

PatentsProsecutionCourtsBuilt from the record

Setup for agents

One connection. Two minutes.

Pick your client. Install Patent Bear from its directory. Sign in with your Patent Bear account. The same steps and screenshots as the Connect page.

Open Settings → Connectors and click Add, then All available. (On a Team plan this is the organization's Connectors page; on an individual plan, your own.)

1 / 7
Step 1 of 7Claude
Claude's Connectors settings with the Add menu open and All available highlighted

Patent Bear is in the Claude connector directory. One connector covers claude.ai in the browser, the desktop app, and Cowork. Nothing to install.

Local models, Cursor, Gemini CLI, and the raw API are on the Connect page.

06 / Measured

Less waiting. Less token waste.

Same ten research prompts, with and without the connector.

up to 12×

faster

about 2.5× on a typical task

up to 86%

fewer output tokens

about half on average

ClaudePatent Bear connected

Elapsed0:00

Pull US 7,844,915, find support for “two or more input points”, then the file history and the court decisions.
  1. ✓Pull the record for US 7,844,9150:04
  2. ✓Claims, parsed and numbered0:06
  3. ✓Find “two or more input points”, with column and line0:11
  4. ✓File history: office actions and responses0:19
  5. ✓Court decisions citing the ’915, to the page0:31

Finished. Every quote cited, in 38 s.

ClaudeNo connector

Elapsed34:23

Pull US 7,844,915, find support for “two or more input points”, then the file history and the court decisions.

    Lookup 44 · still going.

    Measured on 10 research tasks, two rounds, Claude Opus. Small sample. We publish the transcripts. The race above is illustrative.

    07 / Completion

    An answer. Not a dead end.

    Quote the examiner’s § 103 reasoning and the applicant’s answer from the wrapper of US 9,087,321, then save a note. Two agents, same prompt, two endings.

    ClaudePatent Bear connected
    Quote the examiner’s § 103 reasoning and the applicant’s answer from the file wrapper of US 9,087,321, then save a note.
    …get_file_historyUS 9,087,321 · office actions and responses
    …get_file_history_documentNon-Final Rejection · Applicant’s Remarks
    …write_notesaved to the Library

    “Claims 1- 27 are rejected under pre-AIA 35 U.S.C. 103(a) as being unpatentable over Zuckerberg et al. (U.S. 2012/0296749) in view of Svendsen et al. (U.S. 2009/0287763).”Non-Final Rejection of Dec. 17, 2013, quoted from the wrapper as returned.

    “Assigning different point values is not the same thing as generating a likelihood of match between users.”Applicant’s Remarks of June 17, 2014.

    Note saved on US 9,087,321 · 1 min 36 s

    ClaudeNo connector
    Quote the examiner’s § 103 reasoning and the applicant’s answer from the file wrapper of US 9,087,321, then save a note.
    …web_searchPatent Center, Google Patents, public mirrors

    “I could not complete this task, and I want to be precise about why rather than give you something that looks like an answer.”

    “Every public route to the file wrapper failed.”

    “To unblock this: connect the Patent Bear MCP server to this session.”

    Nothing quoted. Nothing saved.

    08 / Security & confidentiality

    Your work
    stays yours.

    Firms run client matters through Patent Bear. Nothing is kept by the AI providers, nothing is used for training, and what you save stays in your account.

    Zero Data Retention.
    Every AI provider we use is bound by signed agreements not to log or keep what we send them.
    No training.
    Your claims, disclosures, and notes are never used to train a model, by us or by our providers.
    Your data, secured.
    Kept only in your Patent Bear account, with row-level security on every record. Deleted on request.

    Connect. Get to work.

    Works in Claude, ChatGPT, Codex, and any MCP client. Two minutes to connect.

    Good afternoon

    How can I help you today?

    +ChatPatent Bear ✓Opus 5.5