Connect PatentDJ to your assistant
PatentDJ speaks the Model Context Protocol, so Claude Code, Claude Desktop and Cursor can search patents, analyse them and write the report without leaving your editor. Searches run against your own credits.
1. Get an API key
Open Account → API keys, name a key, and copy it. The key is shown once and never again — we store only a hash of it, so we cannot recover it for you. Lost one? Revoke it and make another.
2. Point your client at it
Verified against these clients on 4 September 2026. Editors change their config formats; if one of these no longer matches what you see, their own MCP documentation is authoritative and we would like to hear about it.
Claude Code
claude mcp add --transport http patentdj \
https://patentdj.com/api/mcp \
--header "Authorization: Bearer pdj_live_YOUR_KEY"Cursor
Add to ~/.cursor/mcp.json:
{
"mcpServers": {
"patentdj": {
"url": "https://patentdj.com/api/mcp",
"headers": {
"Authorization": "Bearer pdj_live_YOUR_KEY"
}
}
}
}Claude Desktop
Desktop's custom connectors expect OAuth, so a bearer-token server goes through the mcp-remote shim. Add to your claude_desktop_config.json:
{
"mcpServers": {
"patentdj": {
"command": "npx",
"args": [
"-y", "mcp-remote",
"https://patentdj.com/api/mcp",
"--header", "Authorization: Bearer pdj_live_YOUR_KEY"
]
}
}
}3. What you get
| Tool | Cost | What it does |
|---|---|---|
| search_patents | 1 credit | Prose search. Returns a search_id and page 1. |
| search_by_structure | 1 credit | SMILES, exact / substructure / similarity. |
| search_by_sequence | 1 credit | blastp over the patent protein index. |
| analyze_prior_art | included | Risk analysis, 40 patents per call. |
| generate_report | free | Writes the report from the analysis. |
| get_search_results | free | Another page of a search you ran. |
| get_patent | free | Metadata, abstract and citation counts. |
| resolve_structure_name | free | Chemical name to SMILES. |
| get_credits | free | Your balance, and which pool it comes from. |
A typical run: search_patents, then analyze_prior_art repeatedly until it reports remaining: 0, then generate_report. One credit covers all of it.
What this does not cover
- Granted US patents only. No pending applications, no non-US filings. An answer about global freedom to operate is not one this corpus can give.
- Titles and abstracts, not claims. The corpus does not hold claim text, so neither we nor your assistant can tell you what a patent claims. The tools say so in their own descriptions, because the model reading them is the one about to write the sentence.
- Structure search reaches further. The chemistry index is built from patent full text, images and attachments, so a structure match can come from the claims even though the abstract never mentions the compound.
- One search is one credit, and the analysis of that search's results is included. A retry is a second search and a second credit — there is no idempotency key yet, so an agent that retries on timeout will spend twice.
- Rate limits are per key: 40 searches and 120 analysis batches per hour, 240 lookups.
- Search results last 7 days. After that a
search_idstops resolving and the search has to be run again.
Trouble connecting? Email help@patentdj.com.