# GoVeda

> GoVeda is a patent-intelligence platform for AI agents and IP professionals. It offers semantic search over 220M+ patents across 108 jurisdictions (updated weekly), prior-art search, patent records and full texts, patent counts and statistics, company and classification lookup, and AI-generated novelty reports. GoVeda is available as a hosted MCP server with OAuth, a REST API, and a web app at goveda.com.

GoVeda serves two audiences: people using the web app, and AI agents that connect over the Model Context Protocol (MCP) or the OpenAPI REST API. The fastest way for an agent to use GoVeda is the hosted MCP server.

How GoVeda behaves, worth knowing before calling it:

- The MCP server is hosted and managed. There is nothing to install. Claude and ChatGPT users add GoVeda Patent from their assistant's directory; other clients connect to https://mcp.goveda.com/mcp. Either way, sign-in is OAuth in the browser.
- Search is semantic (natural-language descriptions or claims), not keyword or boolean. V2 Search accepts structured per-field filters combined by AND for dates, jurisdiction, language, CPC/IPC, assignees, and legal status; legacy search tools retain their nested search_conditions syntax.
- Coverage is 220M+ patents across 108 jurisdictions, updated weekly. It is not USPTO only. Records include full text, classifications, family, and legal status.
- search_patents, list_patents, and read_patents share one nested patent-record schema: pub describes the publication, app its application, and fam its complete simple family. The three identifiers are included with every selection. Without fields, all three return the same default record: identifiers, title, abstract, key dates, expiration, legal status and current owners. An explicit fields list replaces the default record: records hold the identifiers and the listed fields only, and fields: [] returns only the identifiers. To read claims, description, legal events, transfers, priorities, or family facts, list them together with the default fields you still need.
- Publication-number inputs normalize supported formatting variants and known publication mappings. If a patent tool returns UNRESOLVED_REFERENCE, call discover_patent_references with the original reference. Its candidates are unconfirmed possibilities: continue only after the user selects one or explicitly authorizes choosing a likely match.
- Novelty reports run asynchronously: start one, then poll its status before reading the result.
- V2 record text uses the requested lang preferences, with EN as the default, and reports a fallback when those stored language variants are unavailable. For read_patents with paragraphs, it chooses one language covering all requested numbers, else the first covering any: lang in order, then the original, then any available language. The returned language and is_fallback identify that choice. GoVeda does not machine-translate record text on demand. The deprecated semantic_patent_search keeps its English result summaries.
- Credits are an internal metered unit. Operations cost different amounts, and a report costs far more than a single lookup. Credit spend is expected during normal use.
- Out of scope: general legal advice, trademark search, and copyright questions.

## Connect an AI agent (MCP)

- MCP endpoint (OAuth): https://mcp.goveda.com/mcp
- Claude (claude.ai, Claude Desktop): https://claude.ai/directory/goveda-patent
- ChatGPT: https://chatgpt.com/plugins/plugin_asdk_app_69d32b5b29f88191acf54f9d48afc6e3
- Claude Code: claude mcp add goveda-patent --transport http https://mcp.goveda.com/mcp
- Works with Claude, Claude Code, ChatGPT, Cursor, and other MCP clients.
- [Patent MCP Server](https://www.goveda.com/docs/agent/mcp): per-client setup, OAuth and API-key auth, and what you can do.
- [Tool reference](https://www.goveda.com/docs/agent/mcp-tools): every tool with parameters, return shapes, filter syntax, and per-tool credits.
- [Your first MCP connection](https://www.goveda.com/docs/getting-started/your-first-mcp): connect an assistant and run a first search.

Key MCP tools:

- search_patents: semantic search with v2 filters, collapse policies, exact-term recall and ranked patent records with the default record; other fields as in read_patents
- semantic_patent_search: semantic search across 220M+ patents (deprecated v1; use search_patents)
- search_prior_art, then get_prior_art: find prior art for a published patent or an invention's own text, as a job of a few minutes, read as ranked patent records with the default record; filters and an automatic effective-date cutoff
- prior_art_search: find prior art for a given patent (deprecated v1; use search_prior_art)
- discover_patent_references: find reviewable candidates for an unresolved patent reference without selecting one
- read_patents: read one to 25 patents' fields (text, dates, legal status and events, transfers, priorities, family members and facts) with the same record schema and field catalog as search_patents and list_patents
- list_patent_citations: a patent's backward or forward citations over the publication, its application or its family, a page at a time
- get_patent and batch_get_patents: retrieve patent content by section, single or batched (deprecated v1; use read_patents)
- get_patent_full_text: read one patent's complete abstract, claims, and description as Markdown in a single call
- search_patent_content: find relevant text within one patent
- list_patent_images, get_patent_image, get_patent_image_overview: list a patent's drawings and look at one, or at a contact sheet of many
- get_patent_pdf_page: look at one page of the published PDF to cite its printed column, line, page or paragraph numbers, with a download link to the whole PDF
- get_patent_attachments: download links for a patent's files: the published PDF, drawings and other original files such as chemistry or sequence listings
- get_patent_forward_citations, get_patent_backward_patent_citations, get_patent_backward_npl_citations: citation networks (deprecated v1; use list_patent_citations)
- get_patent_transfers: ownership and assignment history (deprecated v1; use read_patents with app.transfers)
- get_patent_translations: stored multilingual text variants (deprecated v1; use read_patents with lang and family_fill)
- resolve_assignees (free): turn company names in any spelling or script into the canonical assignee names that every assignee filter takes, in search_patents, search_prior_art, aggregate_patents and list_patents
- expand_corporate_group: a company's corporate group (its ultimate owner, subsidiaries, merged companies and former names) as the canonical assignee names those filters take, each with its size
- describe_fields (free): the meaning, kind or type, and values of the v2 fields that filters, breakdowns, orderings and record selections take
- lookup_classifications: find CPC/IPC codes from a description or symbol, including current successors of withdrawn symbols
- lookup_party: a company's canonical names for search filters (deprecated v1; use resolve_assignees, or expand_corporate_group for a corporate group)
- party_count, party_patents: count or list a company's patents (deprecated v1; use aggregate_patents and list_patents)
- aggregate_patents, list_patents: count patents over any slice of the corpus in several units, with rankings, year trends, cross-tabs and company or topic comparisons, and list the patents or families behind a count
- generate_novelty_report, then get_report_status, get_report_summary, get_report_patent_analysis: run and read a novelty and patentability report

## REST API

For direct integration without a Pro subscription. API-key authentication uses a separate, non-expiring API Credit balance rather than the Subscription credits shared by the web app and hosted MCP server.

- [API overview](https://www.goveda.com/docs/api): programmatic semantic search, prior-art search, patent records and statistics (v2), and reports.
- [API keys](https://www.goveda.com/docs/api/api-keys): create and manage keys under Settings.
- [Full API reference](https://openapi.goveda.com/api-references): interactive endpoint reference.

## Web app

- [GoVeda](https://www.goveda.com): semantic patent search and AI reports in the browser.
- [Subscription Pricing](https://www.goveda.com/pricing/subscription): a free OpenPatent tier, OpenPatent Pro in five monthly credit tiers from $200, and Enterprise, with what a month of credits covers in GoVeda Agent tasks.
- [API Pricing](https://www.goveda.com/pricing/api): pay-as-you-go API credits that never expire, with program-year volume discounts.

## Use cases (agent workflows)

Real patent workflows for Claude and ChatGPT through the MCP server. Full set at [Use cases](https://www.goveda.com/use-cases). Highlights:

- [Semantic patent search](https://www.goveda.com/use-cases/semantic-search)
- [Find prior art for a patent](https://www.goveda.com/use-cases/find-prior-art-for-a-known-patent)
- [Generate a novelty report](https://www.goveda.com/use-cases/generate-a-novelty-report)
- [Read a foreign patent in English](https://www.goveda.com/use-cases/read-a-patent-in-english)
- [Explore a company's patents](https://www.goveda.com/use-cases/browse-a-companys-patents)
- [Visualize a competitive landscape](https://www.goveda.com/use-cases/competitive-landscape)

## Docs

- [Getting started](https://www.goveda.com/docs/getting-started)
- [Semantic search](https://www.goveda.com/docs/search/semantic-search)
- [Advanced filters](https://www.goveda.com/docs/search/advanced-filters)
- [Reports](https://www.goveda.com/docs/reports): novelty reports
- [Credits](https://www.goveda.com/docs/plans-and-billing/credits): how credits work and what each operation costs
- [Free trial](https://www.goveda.com/docs/plans-and-billing/free-trial): 14-day Pro trial with 5,000 credits

## Instructions for AI agents

- Use natural-language semantic queries, not boolean strings. Add a filter to narrow by date, jurisdiction, CPC/IPC, or assignee.
- Provide a publication number when retrieving a specific patent, for example US-9939819-B2.
- Read patents with read_patents, or get_patent_full_text for one complete text. To count or rank patents, call aggregate_patents or list_patents.
- Assignee filters take canonical names only: resolve company names with resolve_assignees before filtering search_patents, search_prior_art, aggregate_patents or list_patents by assignee.
- Tools marked deprecated v1 still work but will be removed; use the tool each one names.
- If a patent reference cannot be resolved, call discover_patent_references and present its candidates. Never substitute a candidate unless the user selects it or explicitly authorizes choosing a likely match.
- Reports are asynchronous: call generate_novelty_report, poll get_report_status, then read get_report_summary and get_report_patent_analysis.
- Expect credit usage during normal operation. Do not treat a credit cost as a reason to avoid a needed call.
- Search costs 20 credits per call and List 2 per page, the default record included. Each record that returns a field beyond the default record adds the Read price of 1 credit; the limit reserves an upper bound and the final charge uses the actual count. Read costs 1 credit per returned record, regardless of its field selection. A prior-art search costs 350 credits.
- New accounts get a 14-day free trial with 5,000 credits, which start on the first call.

## Optional

- [Blog](https://www.goveda.com/blog): guides and updates.
- [What is an MCP server?](https://www.goveda.com/blog/what-is-an-mcp-server)
- [Build an AI patent research agent](https://www.goveda.com/blog/mcp-patent-research-agent)
- [Semantic patent search explained](https://www.goveda.com/blog/semantic-patent-search)
- [What's new](https://www.goveda.com/docs/release-notes)
- [Contact](https://www.goveda.com/contact)
- [Privacy](https://www.goveda.com/privacy) and [Terms](https://www.goveda.com/terms)