Tool reference

Natural language search across 220M+ global patents. Describe the technology or invention you’re looking for. Polls internally and returns final results in a single call (typically 5–15 seconds).

Returns: Search results with UCID, title, relevance score, and patent URL for each match.

Find patents similar to a known patent for prior art analysis. Returns a search_id — poll with get_search_status to get results. Typically takes 3–5 minutes.

Returns: search_id for polling with get_search_status.

Check whether a prior_art_search request has completed. When status is completed, the response includes the full results.

Returns: Status and results (when completed) with UCID, relevance score, and patent URL.

Look up CPC/IPC patent classification codes by keyword, technology description, or partial symbol. Useful for finding the right codes to pass into search_conditions for semantic_patent_search or prior_art_search.

Returns: A ranked list of classifications with symbol, title, type (cpc or ipc), level, and a relevance score.

Resolve a free-text company name (e.g., Qualcomm) to the canonical assignee name(s) actually stored on patent records. Required before using the current_assignees / original_assignees filters in search_conditions, because those filters are an exact-string match — Qualcomm, Qualcomm Inc., and QUALCOMM INCORPORATED are not interchangeable, and only one form exists in the index.

Set expand_hierarchy=true to also retrieve the ultimate corporate parent and all subsidiary names. Pass the full all_names list as the value of a current_assignees (or original_assignees) filter to cover an entire corporate group.

Returns: Up to 10 ranked results with name, all_names, relationship (owner, subsidiary, or former_name), ultimate_owner, and country.

Note: Assignee/company names only. Inventor names are not supported — names vary too widely (transliterations, name order, abbreviations) to resolve reliably, and the inventors filter is not available.

Confirm that one or more exact assignee names exist in the index and count how many patents each holds. Deterministic — no LLM, no web; the same input always returns the same output. Use this to verify a portfolio size before listing it, or to compare several companies at once.

Names are matched exactly (including casing) against the stored assignee form — call lookup_party first to resolve a free-text company name. An unknown name is reported per-name (exists: false, patent_count: 0), never as an error, so exists distinguishes an unknown name from a real one filtered to zero.

Returns: results (per-name name / exists / patent_count, in request order), total (union count over patents that have a publication date — a patent held by several listed names counts once), and breakdown (per-authority counts when requested). total equals party_patents’s total under the same names + conditions.

Enumerate the patents held by one or more exact assignee names, publication date descending, as a clean paginated list. Synchronous — no task submission or polling. For semantic / topic search use semantic_patent_search instead; for a free-text company name, resolve it with lookup_party first.

Names are matched exactly. An unknown name simply contributes nothing (no error); a filter matching nothing returns an empty list.

Returns: total (exact count under the same filter), items (patent summaries — publication_number, publication_date, filing_date, authority, patent_status, language, current_assignees, ipcs, cpcs; no title), and next (cursor for the next page, or null when exhausted). Fetch full content per publication_number with get_patent.

Fetch content of a specific patent. You can request specific sections or all of them. Auto-resolves publication number formats.

Returns: Patent content for the requested sections. Description and claims are truncated to 8,000 characters for AI context compatibility.

For citations, use the dedicated get_patent_forward_citations, get_patent_backward_patent_citations, and get_patent_backward_npl_citations tools.

Fetch content for multiple patents in a single request. Same section options as get_patent. Returns partial results if some patents are not found.

Returns: Array of patent content objects. Unfound patents return error objects with the publication number.

Get forward citations — patents that cite this one (patents only, no NPL). Useful for impact / influence and downstream-patent analysis. Forward citations can number in the thousands for foundational patents, so the result is paginated — fetch the next page with the returned pagination.next_offset.

Returns: citations (a list of { ucid, source, title, publication_date }) and pagination (offset / limit / returned / total / has_more / next_offset).

Get backward patent citations — patents that this patent cites (patents only; for non-patent literature use get_patent_backward_npl_citations). Each item carries the examiner relevance category (X / Y / A …) and rel_claims (which claims of this patent the reference is cited against) where available. Paginated — fetch more with pagination.next_offset.

Returns: citations (a list of { ucid, source, category, rel_claims, title, publication_date }) and pagination.

Get backward non-patent-literature (NPL) citations — papers, books, standards, etc. that this patent cites. NPL exists only on the backward direction. Paginated — fetch more with pagination.next_offset.

Returns: citations (a list of { text, source, category, author, title, date }) and pagination.

List the stored binary attachments of a patent — the full-document PDF and drawings (figures). Returns metadata and a download_url for each file. Binary content is not returned over MCP — download from the URL with your X-API-Key header. Coverage varies by authority and publication date; an empty list means nothing is stored.

This tool is behind a feature flag. If it does not appear in your client’s tool list, it is not enabled for your account — use the REST endpoints below instead.

Returns: count and an attachments array, each with filename, media, size, kind, and download_url.

Convenience REST endpoints (outside MCP): GET /api/patents/{publication_number}/pdf (full PDF) and GET /api/patents/{publication_number}/images/abstract (cover figure as PNG).

Get a patent’s transfer (assignment) history — who owned it and when ownership changed. Combines USPTO reassignment records (US patents) and transfer legal events (CN/EP/… patents) into one timeline, newest first. For the current owner only, use get_patent with sections=parties instead. An empty list means no transfer was recorded.

Returns: count and a transfers array, each with date, type, from_parties, to_parties, details, and source.

Get the language variants of a patent’s text (title, abstract, claims, description) as stored in the corpus — for example, Chinese patents carry both ZH and EN text. This is not on-demand machine translation: if the requested language is not stored, the best available variant is returned with is_fallback=true — check available_languages in the response.

Returns: The requested fields plus available_languages. Long fields are truncated to 8,000 characters.

Run a novelty and patentability assessment for an invention description. Returns a report_id — poll with get_report_status to track progress. Typically takes 5–15 minutes.

Returns: report_id for polling with get_report_status.

Check the progress of a novelty report. Returns stage information and patent analysis counts.

Returns: Progress with stage, total/processed/relevant patent counts, and elapsed time.

Stages: pending → searching → analyzing_query → analyzing_patents → analyzing_final → completed

Get the patentability assessment and a directory of analyzed patents from a completed report. Report status must be completed.

Returns: Patentability assessment, executive_summary (verdict explanation), relevant_prior_art (top threats deduped from most_concerning_patents + key_threats + relevance-ranked patents), and patent directory (up to 30 items sorted by relevance score).

Retrieve detailed per-patent analysis from a completed report. Report status must be completed.

Returns: Detailed analysis for each patent with relevance scores, sorted by relevance.

Returns your current credit balance and billing period details.

Not available in ChatGPT. This tool is exposed to Claude, Cursor, and API-key clients only. In ChatGPT, check your balance on the Plan & Credits  page instead.

Returns: Total remaining credits, reserved credits (held for in-progress operations), and billing period details with expiry dates.