Inspiration
AI agents can perform advanced scientific computations, but researchers demand transparent pricing, explicit consent, and verified payments before any paid workflow executes.
MatterVis AI brings those controls into a scientific workspace, connecting an interactive crystal viewer with an assistant that can explain structures and arrange paid analysis around a researcher's choices.
What it does
Researchers inspect structures and build supercells for free. For disorder analysis, the MCP tool presents the method, candidate count, source checks and USD 2.00 PayPal Sandbox price, then asks for explicit consent. PayPal buyer approval is a separate step. The server verifies the approved order and capture before the calculation begins; the transaction uses virtual Sandbox funds.
Caffeine demonstration
The bundled example is the caffeine crystal COD 4504310. Its formula-unit view shows part of the loaded structure, while disorder occurs at other sites. Switching to the unit-cell view reveals overlapping alternative components and their clash warnings. Displayed counts include copies.
After the verified capture, MatterVis generates and checks up to two enumerated candidates. Researchers can open each structure, review the checks and limitations, and download a results package with provenance. The candidates are hypotheses for human review. Establishing physical stability and the experimentally dominant configuration requires independent validation.
How we built it
MatterVis supplies the viewer and MolCrysKit performs structure inspection and candidate calculations. A real MCP client and server connect the assistant to these operations, with MCP elicitation for human consent. A frozen server-side quote binds the structure, method, candidate count and price to a PayPal Sandbox Orders v2 transaction. The server checks the order and capture before unlocking the paid workflow.
Challenges and what we learned
Alternative disorder sites can overlap in the deposited model and generate clash warnings. Showing the source warnings alongside candidate checks gives researchers the context to interpret each result. The payment workflow combines a clear quote, a human decision, buyer approval and server verification with an asynchronous scientific calculation.
What's next
Expand evaluation across additional disorder types and work with researchers on candidate review and provenance.
Data and code
The caffeine structure comes from Enright, Terskikh, Brouwer and Ripmeester, Crystal Growth & Design 7 (2007), 1406–1410, doi:10.1021/cg070291o. COD data are dedicated to the public domain under CC0.
Source code, setup and tests: https://github.com/SchrodingersCattt/mattervis-saas-paypal
