Model Context Protocol (MCP)
Integrating Thinkstruct with your Agents
Thinkstruct exposes an MCP server so AI agents such as ChatGPT and Claude can run patent analysis on your behalf. Connect it once, then start searches, read and interpret the results, and open the full work in your Thinkstruct workspace, all from inside the conversation.
What you can ask
No commands to learn. Describe the analysis you want and your agent handles the rest. It can run invalidity and prior art searches, freedom to operate assessments, and infringement searches, then read the results with you, rank the references, build claim maps, and link you into your Thinkstruct workspace. Paste a Thinkstruct link and it pulls that session back in to summarize what the search found.
-
Run an invalidity search on
US10000000B2, then walk me through the three strongest references and why they read on claim 1. - Do an FTO assessment for this product description and tell me which patents look most blocking.
- Build a claim map for the top prior art hit and summarize where the overlap is strongest.
- Here is a Thinkstruct session link. Pull it in, summarize what the search found, and tell me what to look at next.
Capabilities & tools
The connector exposes the tools below. Your agent chooses which to call based on what you ask; you never call them directly. A few behaviors are worth understanding.
How it works
- Searches are asynchronous. Search and claim-map tools return a job id immediately; the agent polls until the job is done, then reads the results. This can take from seconds to a few minutes.
- Work is organized into sessions. A session is simply a place to store the searches and analysis about a single patent. The agent prefers to attach a search to a session and save the results back to it, so the analysis is persisted and you can open it in the Thinkstruct app. Related sessions can be grouped into a portfolio.
- Results are shown as claim charts. Claim-mapping results are always presented as a claim chart, mapping each claim limitation to where a reference discloses it.
- Links, not identifiers. Where an item has no readable name, the agent gives you a link to open it in Thinkstruct rather than an internal id.
Search & analysis
| Capability | What it does | Credits |
|---|---|---|
| Invalidity / prior art search | Find prior art that may invalidate a patent's claims. | 90 |
| Infringement search | Find products or patents that may infringe a patent. | 100 |
| Freedom to operate (FTO) | Find patents that could block a product or invention. | 60 |
| Patentability search | Find prior art relevant to whether an invention is patentable. | 80 |
| Build a claim map | Generate a custom claim chart for a reference: the agent unlocks the reference in a session and queues the claim mapping. | 10 |
Reading & organizing
| Capability | What it does | Credits |
|---|---|---|
| Look up a patent | Get a patent's title, abstract, and claims by number. | Free |
| Find patents by assignee | List patents owned by a company. | Free |
| List & open sessions | Browse saved analyses and open any in the app. | Free |
| Read a session | Summarize a session's results and claim charts (paste a link or pick one). | Free |
| Portfolios & recent activity | Browse portfolios and recently accessed items. | Free |
| Resolve references | Turn reference identifiers into readable titles. | Free |
| Check credit balance | See remaining credits before a search. | Free |
Creating
| Capability | What it does | Credits |
|---|---|---|
| Create a session | Start a new analysis for a patent (litigation, patentability, or FTO). | Free |
| Create a portfolio | Make a named collection to group related sessions. | Free |
| Add a session to a portfolio | Organize a session into one of your portfolios. | Free |
| Upload a text document | Add a product description or write-up to use as a reference. | Free |
Connect an agent
Add Thinkstruct as a connector in your agent using this URL, then follow the steps for your agent below.
Custom connectors require a ChatGPT Business, Enterprise, or Edu plan (Pro/Plus support is limited). You also need to enable Developer Mode under Settings → Apps & Connectors → Advanced.
- Enable Developer Mode in Settings, Apps & Connectors, Advanced.
- In Apps & Connectors, add a custom connector and paste the Thinkstruct URL above.
- Sign in with your Thinkstruct account and approve access on the consent screen.
- Start a new chat, enable Thinkstruct in the tools menu, and ask it to run a search.
- Open Settings, then Connectors.
- Choose Add custom connector and paste the Thinkstruct URL.
- Sign in with your Thinkstruct account and approve access on the consent screen.
- Enable the Thinkstruct connector in a conversation and ask it to run a search.
Custom connectors are available on Claude's paid plans.
- Open Settings, then the MCP section.
- Add a new MCP server and enter the Thinkstruct URL as a remote server.
- Complete the Thinkstruct sign in when prompted.
- Thinkstruct tools become available in chat and composer.
Cursor supports remote MCP servers with OAuth sign in.
- Run the Add MCP Server command from the Command Palette.
- Choose an HTTP server and enter the Thinkstruct URL above.
- Sign in with Thinkstruct when the browser prompt appears.
- Open Chat in agent mode to use the Thinkstruct tools.
Use VS Code's agent mode to call the tools.
- Find where your agent adds a remote or HTTP MCP server.
- Enter the Thinkstruct URL above as the server address.
- Complete the Thinkstruct sign in and approve access.
- The Thinkstruct tools appear in the agent's tool list.
Any agent that supports remote MCP servers with OAuth will work.
Sign in & credits
When you connect, you are sent to a Thinkstruct sign in and consent screen. Your agent never sees your password, and you can revoke access at any time from your account settings. Searches run through an agent use the same credits as searches on the site, so your agent shows the cost before it runs and your remaining balance after. Reading and interpreting results does not cost additional credits.
A note on your AI agent Thinkstruct is not responsible for the security or the quality of the AI agent you connect. Our information retrieval models are built so that they cannot hallucinate, and we stand behind the accuracy of the results Thinkstruct returns. We cannot, however, guarantee the output of whatever model you use to read or summarize those results. Always confirm an agent's summary against the underlying results in your Thinkstruct workspace.
Trouble connecting? Contact support.
Example usage
A complete invalidity search inside Claude, from kickoff to a saved session and an exported claim chart.
Done — the search completed and results are saved to a new litigation session. Here's the rundown:
Focal patent: US9876543B2 — "Adaptive beamforming for multi-user MIMO wireless systems" (Aterlon Communications)
Top references found (30 total):
- US8123456B2 — Downlink beamforming with per-user channel feedback (Qualcomm) — strongest match, claim chart generated
- US8654321B1 — Codebook-based precoding for MU-MIMO (Samsung)
- US9011223B2 — Adaptive antenna weighting from CSI reports (Ericsson)
- US8790011B2 — Interference-aware user grouping (Huawei)
- US9200455B1 — Pilot allocation for channel estimation (NTT DoCoMo)
Claim chart — US8123456B2 vs US9876543B2:
| Focal claim / limitation | Where US8123456B2 discloses it |
|---|---|
| Claim 1, limitation 1 | Abstract — a base station computing beamforming weights from per-user channel state feedback across an antenna array |
| Claim 1, limitation 2 | Col. 4:15–4:33 — selecting a precoding matrix per scheduled user |
| Claim 1, limitation 3 | Col. 6:2–6:28 — updating the weights as CSI reports arrive, matching the adaptive-update limitation |
| Claim 12, limitation 2 | Fig. 3 & Col. 8:40–8:57 — feedback loop for weight refinement |
View the full results and claim charts here: https://thinkstruct.com/compare?session=e2918bdf-1bc2-4ac6-a486-b27a5da74a2b
Something on the Thinkstruct website your agent can't do yet, or is your agent having trouble with a task? Let us know and we'll help.