What Juristat's MCP server actually is
Juristat runs an MCP server that exposes its patent data of more than 10 million pending, abandoned, and granted USPTO applications, plus examiner, firm, and office action analytics as a fixed set of "callable" tools than can be accessed as a source when a user asks a question using an AI Large Language Model (LLM) like Claude, CoPilot, or ChatGPT. It doesn't matter which LLM someone is using, one of the aforementioned frontier LLMs or any other MCP-aware assistant; MCPs connect to the server the same way, over a shared JSON-RPC interface.
The LLM application someone is working in runs the MCP client side of that connection. It talks to Juristat's server, asks what's available, and carries requests back and forth between the model and Juristat's data.
The LLM and Juristat's MCP server do different jobs
The LLM
Reasons over language. It reads a question like "how is this examiner trending on §101?" and decides a live lookup is needed. It has no memory of Juristat's data on its own; every number it gives back has to come from somewhere else.
Juristat's MCP server
Does the actual work of providing source information. It runs the examiner report, the portfolio query, or the office action search against Juristat's own database and returns the result. It is isolated from what the rest of the conversation was about and only sees the one request it's handed.
How the model picks a Juristat tool
Before an answer is provided by the LLM, the model is given a manifest of every tool that Juristat's server exposes (e.g. searching applications, pulling an examiner's report, checking office action metrics), each with a description and the input it expects. The model matches the question against those descriptions and emits a request naming one tool and filling in its arguments. Only then does the host route that request to Juristat's server and anything actually runs.
question about an application or examiner → Juristat's tool manifest (search-applications, get-examiner-report, ...) → model picks a tool and fills its arguments → Juristat's server runs it
What actually gets sent to Juristat
The model doesn't hand Juristat the conversation. It hands over exactly one thing: the tool it picked, and the arguments it generated for that call.
No system prompt, no chat history, no other client's data — just this one call. Whatever Juristat sends back (the matching applications, the examiner's numbers) becomes new context for the model's next turn, not a standing connection into Juristat's database.
To learn more about what users can do with Juristat's MCP, visit this page, or you can learn more about Juristat's security and privacy policies here.