Getting a part quoted at a print bureau usually works the same way everywhere. You open a browser tab, upload the file, choose a material and wait for the calculator to give you a price. Romanian service bureau 3Descu now offers a second way in. On October 18, 2026 it published a Model Context Protocol server that lets AI assistants such as ChatGPT, Claude, Cursor and VS Code ask its quoting system for prices. The server appears in the official MCP Registry as io.github.cadi27tech/3descu-quote, titled "3Descu 3D Printing Quotes (EU)." It is listed at version 1.0.0, it is reached as a streamable-http remote endpoint, and its status has been "active" since the launch date.
The registry describes the service as "Instant FDM and SLA 3D printing quotes from an ISO 9001:2015 certified EU factory, 200+ printers," and it points to quote.3descu.com as the related website. 3D Printing Industry reported the launch on October 22 with more detail on what the tools do and, just as important, what they don't do.
What the Server Actually Exposes
The interface is small. 3D Printing Industry reports four functions:
- Capabilities: what the bureau can produce.
- Materials: the materials it offers for its FDM and SLA processes.
- Model analysis: a geometry check that measures a mesh's volume, bounding dimensions and surface area, and reports whether the mesh is watertight.
- Price estimate: the quote itself.
Splitting the work this way makes sense. Capabilities and materials give an assistant the context it needs before asking for a price. According to the report, the analysis measurements are then used to generate a price estimate based on the selected material and quantity. A non-watertight mesh is the classic file that breaks slicers and pricing tools alike. Checking for it up front means the assistant can report the problem in the conversation instead of returning a price that rests on bad geometry.
According to the report, the server runs on the same calculation engine as 3Descu's website. A quote from an assistant should therefore match a quote from the browser upload. That matters because the obvious worry with any AI-assisted pricing is that the number came from a language model's guess rather than from the vendor's own rules. In this design the assistant only passes the request along. The arithmetic happens in 3Descu's existing system.
How You Feed It a Model
The server doesn't take file uploads directly. For model analysis it needs a publicly accessible HTTPS link to an STL or OBJ file hosted on a supported domain. In practice your model has to be hosted somewhere the service can fetch it, and the report does not list which domains are supported, so check that your host works before relying on it. The assistant passes the URL to the quote tools, and the service analyzes the mesh. You also choose a material and a quantity.
For many makers this is a minor annoyance. For anyone with confidential geometry it is a real factor. A publicly reachable link is, by definition, reachable by others who have it. If you're quoting a client's proprietary bracket rather than a printable you've already published, think about where you host the file and for how long before you paste the link into a chat.
The setup itself has no friction. The tools need no account and no API key. You add the remote server to an MCP-capable client and start asking questions.
Where the Assistant Stops
The most important design choice may be what 3Descu left out. You can't place an order through the AI assistant. The tools return a link to a quote page on 3Descu's site, where the customer continues the process. 3D Printing Industry also reports that the production team reviews every file before manufacturing begins, and that, according to the company's online quotation documentation, pricing may be adjusted if technical issues are found in that review, including problems caused by mesh geometry or requirements that need manual assessment.
That second point is worth noticing. The automated estimate is a starting figure, not a binding price. Human review is a sensible step at a production bureau, and keeping it means the agentic layer speeds up the "roughly how much?" stage without letting software commit the shop to a job.
It also limits the damage an assistant can do. Misreading a request, picking the wrong material or quoting the wrong file costs nothing worse than a wrong number on screen. No charge and no print job can follow from a misunderstanding in a chat window.
Who Is Behind It
3Descu runs from Târgu Jiu, Romania. The trade report describes an ISO 9001:2015-certified facility running more than 200 FDM and SLA printers. The company says it has completed more than 1,200 orders for over 450 clients. That makes it an established mid-sized bureau, not a large contract manufacturer, and the MCP launch looks like a way to show up in a new place where engineers and makers already work. Founder Adrian Căpraru is quoted in the 3D Printing Industry piece: "Engineers already work with AI assistants every day. Now they can get a real price from a real factory in the same chat, without filling in a single form."
Cursor and VS Code on the list of supported clients is a meaningful detail. A developer or engineer writing code next to a CAD export can get a production estimate without leaving the editor. ChatGPT and Claude cover the more conversational case, such as "here's my enclosure, what would it cost in SLA versus FDM?"
What It Means for Makers
For hobbyists who print everything on their own machines, this changes little. For makers who send out the parts their home printers can't handle well, such as large FDM pieces, fine-detail resin parts or small batches, it offers a quicker way to compare processes and materials before going to a quote page.
Some practical takeaways:
- Treat the number as an estimate. The company's documentation says pricing may be adjusted after its review if technical issues turn up.
- Use the analysis tool even if you never order. A quick watertightness check plus volume and size readouts is useful in itself before you send a file anywhere.
- Mind the hosting requirement. The public HTTPS link requirement is the main trade-off. Don't expose files you aren't comfortable sharing.
- Expect to finish in a browser. The assistant gives you a link, not an order confirmation.
More broadly, 3Descu is an early example of a print bureau treating AI assistants as a front end for quoting. It keeps the bureau's own pricing engine and its staff in charge of what actually gets made. Whether other bureaus follow will depend on whether quotes requested through an assistant turn into real orders. As a first version it is well thought out: the assistant is limited to asking for quotes, and placing the order and approving production stay with the customer and the factory.
