Patent Intelligence API

Patent data infrastructure for builders.

Programmatic access to the same AI-powered patent intelligence that drives PatentNest — hybrid semantic search, AI feature extraction, and element-wise novelty evidence mapping. REST endpoints and MCP tools, authenticated with one API key, metered with transparent quotas. Starting with comprehensive Indian patent coverage, with additional jurisdictions on the roadmap.

Request a free trialTalk to us

Search coverage manifest

returned with every query
Corpusindian-patent-journal
JurisdictionIN
Documents indexed160,000+
Semantic coverage99%+
Search modehybrid-semantic-text
Embedding modeltext-embedding-3-small

Every search response includes this manifest. When the corpus grows or a new jurisdiction comes online, your integration sees the change in the response — no polling, no changelog to watch.

Element-wise evidence verdict

per feature × per patent
Phase-change thermal storage panelpresent
Photovoltaic charging of storagepartial
Predictive pre-cooling controllerabsent
Door-opening schedule integrationunknown

Each verdict carries a verbatim quote and the field it was found in. An “unknown” means insufficient evidence in the available text — not “probably novel”. Quality flags report when the evidence basis is thin.

Coverage

160,000+ Indian patents

Search

Hybrid semantic + full-text

AI analysis

Feature extraction & evidence mapping

Protocols

REST + MCP for AI agents

Use cases

What people build on top of it.

The API is used by legal tech platforms, IP analytics tools, R&D workflow automation, and AI agents that need patent intelligence as a building block — not a destination.

Patent landscape analytics

Build technology maps, monitor competitive filings, and track assignee activity across the Indian patent jurisdiction — programmatically, on your own schedule.

Prior art screening

Integrate prior-art screening into your invention intake workflow. Submit disclosures, get structured features and ranked prior art — before a human analyst touches the case.

IP due diligence

Feed feature-level evidence into M&A, licensing, and freedom-to-operate workflows. Every verdict carries a verbatim quote and source field — ready for citation.

Novelty assessment automation

Run the full extract → search → map pipeline from code. Build element-wise novelty matrices that compare each invention feature against multiple prior-art references.

AI agent workflows

Connect Claude, Cursor, or your own agent stack via MCP. Agents autonomously research prior art, evaluate patentability, or generate structured reports — with the same data and quotas as REST.

Legal tech platforms

Embed patent intelligence into case management, prosecution tracking, or filing systems. Structured JSON responses with stable schemas, versioned endpoints, and predictable rate-limit headers.

Workflow

Three calls. Disclosure in, evidence matrix out.

The API is designed around how novelty analysis actually works: decompose the invention, find what exists, then build an element-wise evidence chart. Each call produces structured output that feeds directly into the next — no reformatting, no copy-paste between tools.

  1. 1

    Describe the invention in plain English

    extract

    Send a disclosure (40–20,000 characters) to the feature extraction endpoint. It runs the same AI normalization stage as the full PatentNest novelty pipeline — atomic technical features, novelty-focus candidates, a ready-to-use prior-art search query, and CPC/IPC classification hints. No patent vocabulary required from your side.

  2. 2

    Search the corpus with the suggested query

    search

    Use the AI-suggested query — or write your own — to run hybrid semantic and full-text search across the patent corpus. Results come back ranked with composite, semantic, and text relevance scores. Every response includes a coverage manifest: corpus, jurisdiction, document count, and the share that is semantically indexed. A negative result is a measured signal, not a guess.

  3. 3

    Map features against shortlisted patents

    map

    For each shortlisted result, submit the features and the publication number. Each feature is classified as present, partial, absent, or unknown — with a verbatim quote from the patent text and the field it was found in (title, abstract, or claims). You get the evidence, not a summary of it.

Endpoints

Four endpoints. No SDK required.

Standard REST with JSON request/response and bearer auth. Every endpoint returns structured data with a request ID, duration, and rate-limit headers. Errors use stable, machine-readable codes. Import the OpenAPI 3.1 spec into Postman, Swagger UI, or any code generator.

Semantic patent search

passes
POST /api/v1/patents/search

Hybrid semantic and full-text search across the patent corpus. Send a plain-English technical description (2–2,000 chars), get up to 50 ranked patent records with relevance scoring and a coverage manifest that declares what was searched.

Patent record lookup

passes
GET /api/v1/patents/{publicationNumber}

Retrieve a single structured patent record by publication number. Case and separators are normalized automatically — any reasonable formatting resolves to the same canonical record.

AI feature extraction

passes
POST /api/v1/analysis/features

AI analysis of a plain-English invention disclosure. Returns atomic technical features with per-feature confidence and type classification, novelty-focus candidates, a retrieval-optimized search query, and CPC/IPC classification hints.

Element-wise evidence mapping

passes
POST /api/v1/analysis/feature-mapping

Classifies each submitted feature (1–12) as present, partial, absent, or unknown against one patent record — with a verbatim quote and the field it was sourced from. Returns quality flags for low evidence, ambiguous abstracts, and language mismatch.

Data

Structured records, not summaries.

Every patent record includes applicants as arrays with addresses, dates in ISO 8601, scores decomposed into semantic and text components, classification codes for landscape analysis, and source provenance back to the original government publication.

Publication & application numbers

Canonical publication number, raw application number, country code, and kind code. Normalized and deduplicated across journal issues.

Title, abstract, and full text

Complete title and abstract as published. Full claims text available for AI analysis endpoints (feature mapping evidence is quoted verbatim from it).

Structured applicants

Array of applicant objects — name, address, and sequence number. Not a concatenated string. Ready for assignee landscape analysis.

Inventors and classifications

Inventor list as published. IPC and CPC classification codes for technology landscape mapping, trend analysis, and cross-jurisdiction comparison.

Filing and publication dates

ISO 8601 dates. Build prosecution timelines, calculate pendency, filter by filing window, or track publication velocity.

Extraction confidence and source provenance

Machine-readable confidence score on data extraction quality. Every record carries the original source document name and page number — full audit trail to the government publication.

Relevance scoring (search)

Composite score, separate semantic and text-match scores, and the list of fields that contributed to the match. The ranking is transparent about why each result placed where it did.

Coverage manifest (search)

Corpus name, jurisdiction, total document count, semantic coverage percentage, search mode, and embedding model. Returned with every search so your integration always knows what was searched.

AI agents

Native MCP support for the agent era.

POST /api/v1/mcp exposes all four operations as MCP tools over streamable HTTP. Same bearer keys, same quotas, same structured data — but now Claude, Cursor, or your own agent stack calls them directly without glue code. Sessions are stateless. Protocol versions 2025-06-18, 2025-03-26, and 2024-11-05 are supported.

Claude Desktop and Claude Code

Add the MCP config to your Claude settings. Four patent tools appear alongside your other tools. Your agent searches prior art, extracts features, and builds evidence charts in a single conversation.

Cursor, Windsurf, and AI IDEs

Any IDE that speaks MCP can connect. Patent intelligence becomes part of a coding or research workflow without leaving the editor.

Custom agent platforms

Standard JSON-RPC over streamable HTTP. Build agents that autonomously screen invention disclosures, monitor competitive filings, or generate prior-art reports.

REST for everything else

No agent framework needed. Standard HTTP in any language. The MCP transport is there for tools that speak it — everyone else uses the four REST endpoints.

MCP server configuration

{
  "mcpServers": {
    "patentnest": {
      "type": "http",
      "url": "https://patentnest.ai/api/v1/mcp",
      "headers": {
        "Authorization": "Bearer pn_live_your_key"
      }
    }
  }
}

Trust and limits

Transparent metering. Your disclosures stay yours.

Rate limits are per-minute, daily, and monthly — all returned in every response header. AI analysis has a separate daily budget. Rejected requests never consume a credit. Submitted invention disclosures are processed and forgotten — never stored as text, never used for training.

Request quotas

Per-minute, daily, and monthly budgets per client. Headers on every response: RateLimit-Limit, RateLimit-Remaining, X-RateLimit-Daily-Remaining, X-RateLimit-Monthly-Remaining. HTTP 429 includes Retry-After.

Analysis budget

The two AI analysis endpoints run full LLM inference — they have their own daily credit pool. A credit is charged only after validation passes and the model starts. A 400 or 404 costs nothing.

Data confidentiality

Submitted disclosures produce the response, then are not retained. Request logs store hashes and metadata, never invention text. No model training on your data. Request bodies capped at 256 KB.

Coverage

Starting with India. Built to grow.

The Indian patent corpus — sourced from IP India Patent Journal publications — is the first jurisdiction on the platform. The same API surface, data schema, and query format will extend to additional patent offices as they come online. Integrations built today won't need to change.

Indian Patent Journal corpus

160,000+ patent records with 99%+ semantic embedding coverage. Full text extraction from published journal PDFs, deduplicated across issues, with source provenance on every record.

Jurisdiction roadmap

Additional jurisdictions — including EPO and USPTO — are in development. The API schema and endpoint structure are jurisdiction-agnostic by design, so new corpora appear as additional search coverage, not new endpoints to integrate.

Documentation

Complete reference, OpenAPI spec, and quick start.

API reference documentationOpenAPI 3.1 specification

Get API access for your platform.

API keys are provisioned per client with configurable rate limits. Tell us what you're building and we'll set up access, walk through the endpoints, or provision a sandbox environment for your integration.

Request a free trialTalk to us

Trial requests are reviewed by a person, usually within one business day. No card, no auto-renewal.