For AI agents

An MCP server for manufacturing: real answers for your agent.

A language model can write a quote request in seconds. It can't tell you whether a feature can be made, how the part would be held or how long it would run. Connect it to Protyus and it can: your agent uploads a STEP file, runs the analysis and reads back the setups, DFM findings, cycle time and cost as structured data with units.

How it connects

One address, an API key and your organization.

The MCP server is at https://app.protyus.ai/api/v1/mcp, over Streamable HTTP, with one tool for each API operation. Make an API key in the app under Preferences, and send it with your organization's id.

Building your own integration? Every tool is also a REST operation, described in the OpenAPI document. The API and MCP server come with Pro and Enterprise, and with Pro's free trial.

Connect Claude CodeStreamable HTTP
claude mcp add --transport http protyus \
  https://app.protyus.ai/api/v1/mcp \
  --header "Authorization: Bearer ak_YOUR_KEY" \
  --header "X-Protyus-Org: org_YOUR_ORG"
Then ask

“Analyze bracket.step with Protyus in my project and tell me its setups.”

In a conversation

Ask for a part. Get what was measured.

It calls the tools
Asked for a part, the agent uploads the file, starts the analysis and waits for the result.
It reads what was measured
Setups, findings, cycle time and cost come back as data, so the agent quotes the numbers and doesn't guess them.
It shows the part
A finding comes back as a picture of the part, with what it points at numbered.
Your agentProtyus MCP
Analyze Clamp_1.STEP with Protyus and tell me its setups.
create_uploadstart_analysisget_analysis

Clamp_1.STEP can be made as modeled, and scores 77 out of 100.

On the Haas UMC-500 it takes two setups. The first reaches four sides of the part.

What would a shop push back on? Show me.
get_analysisview_part_picture

Two things:

  • The slot (1). It's 3 mm wide and 60 mm deep, too deep for an end mill, so it needs a slitting saw.
  • The R3 corners. They match a Ø6 cutter exactly. Go slightly larger so the tool cuts them on an arc.
3 mm slot, 60 mm deep: cut with a slitting saw
What will 50 cost?
get_analysis

$69.94 each, $3,496.94 for 50, in 6061-T6, at 11.8 min a part.

Ask your agent
Enter to send
What an agent can do

From a STEP file to a quote, as tools.

Analyze a part
Upload a STEP file, pick the process, machine, material and quantity, and run the analysis.
Read the plan
Setups, DFM findings, cycle time and cost, as structured data with units, a section at a time.
Show the part
A finding, a setup or the whole score shown on the part, as a picture or a view the person can turn.
Compare iterations
Upload a changed file of the same part and read what changed: the findings fixed and added, and the score, time and cost.
Price and quote
Price an analysis, keep the price as a quote, and get the PDF to send the customer.
Work with the shop's data
Read and change the machines, materials, cutting tools, tolerances and settings every analysis runs on.
Built for agents

Answers an agent can rely on.

The same permissions as the app
An agent acts as the person whose API key it holds, in their organization, and can do exactly what they can.
The same answer every time
Run a part twice with the same settings and the answer is the same, to the last number. An agent can rely on it.
Marked for safety
Each tool says whether it only reads, changes the organization's data, deletes something or reaches someone outside, so a client can ask first.
Errors an agent can act on
Every error has a stable code, a message and, when one argument is at fault, its name.
Read next

AI DFM

Why a language model can't review a part on its own, and what it can do with a measured one.

DFM analysis

What a DFM check finds, how findings are ranked, and how it leads to a cycle time and a cost.

Questions

What people ask.

Anything else? Email us.

What is an MCP server?

The Model Context Protocol (MCP) is an open standard that lets an AI assistant use outside tools. An MCP server gives the assistant a set of tools to call. Protyus's gives it manufacturing analysis: upload a part, run an analysis and read back the setups, findings, cycle time and cost.

Which AI clients work with Protyus?

Claude, Claude Code, ChatGPT, Codex, Microsoft Copilot, Cursor and other MCP clients. They connect over Streamable HTTP with an API key from the app and your organization's id.

Is there a REST API as well?

Yes. Every MCP tool is also a REST API operation, described in a public OpenAPI document. Calls take an API key from the app as a bearer token.

Which plans include the API and MCP server?

Pro and Enterprise, and Pro's free trial. On other plans a key is refused.

What can an agent change?

Only what the person who made the key can change in the app. Each tool says whether it only reads or changes something, so a client can ask the person before it creates a quote, edits a machine or sends an email.

Why does an AI agent need this?

A language model can't measure a part, so on its own it guesses what can be made and what it costs. Protyus works those out from the part's geometry, on your equipment and at your rates, and the agent builds on the result.

Give your agent a part it can measure.

Protyus is opening to shops, suppliers and design teams by invitation. Ask for early access, or write to us with any question.