Ask Me Anything: Introducing Epsilon3 MCP Preview
A rocket program generates some of the most valuable data in all of science and engineering. Period. Every procedure, every run, every signoff, every part and tracking number, every open issue, it’s a complex system of records that can become overwhelming sometimes. And almost all of it lives one or two clicks too far away from the person who needs it.
Let’s say you are mid-run and want the components and quantities for the assembly in front of you. You are writing a readiness review and need every run that started last month. You are picking up someone else's issue and want to know who touched it and when. The answers exist, but getting to them means stopping what you are doing, opening another screen, and reconstructing context across multiple screens.
Today, we’re closing that gap with our new Epsilon3 Model Context Protocol (MCP) server, our latest efforts to help you leverage your Epsilon3 data in ways that drive your operations forward. It lets an AI assistant you may already use (Claude, ChatGPT/Codex, etc.) connect securely to your Epsilon3 environment and answer questions about your live Epsilon3 data, in plain language, without you clicking through a single screen. We have released it in a beta preview for users on an opt-in basis.
What an MCP server actually is
MCP is an open standard for connecting AI assistants to real systems. Think of it as a secure adapter between your assistant and Epsilon3. Once you connect it, the assistant can look things up on your behalf: search your procedures and runs, open a record in full, query issues, check inventory, and summarize what’s happening across the workspace.
You do not learn an API. You do not write a query. You ask a question the way you would ask a teammate, and the assistant figures out which tool to call and how to call it.
Under the hood it is a small, deliberate set of MCP read tools:
Search across procedures, runs, parts, and issues.
Open any one of those records in full detail.
Summarize the operational picture: what is running, what is open.
Query issues by status and by what they are linked to.
Check inventory for a part by name, ID, or tracking number.
Find Related Entity Given a part, procedure, or run, find all directly related E3 entities.
WhoAmI answers who the current logged in user is, for querying about “my runs” or “my issues”.
Six read-only tools, one conversation.
What you can actually ask
Find a run or runs
“What were my runs in the last 7 days?”
“What ultimate test runs did Adam run within the last 2 months?”
The assistant uses MCP tools to navigate participant filtering and can find your runs, or a colleague’s runs so you can find that particular tricky run you did 2 months ago.
Find a procedure and read it in full.
"Find the QA acceptance procedure and show me its full steps."
The assistant resolves the right record from a name or a code, then opens it completely. Every section, step, and telemetry check comes back, with nothing truncated even on the largest procedures.
Compare two procedures side by side.
"Compare the two takeoff procedures and show me where they differ."
It pulls both in full, so you can line up sections and steps without flipping between tabs.
Read a procedure's telemetry checks.
"Which steps here verify a telemetry parameter, and what limit does each one check?"
Telemetry steps come back human-readable, each with its parameter name, units, rule, and value.
Work through your issues.
"Show me the open issues, then the resolved ones."
"List every open issue, all of them."
Status filtering behaves exactly as written, and a list request returns the complete set, not a truncated first page.
Trace what is linked.
"Which issues are linked to this run?"
"Who commented on this issue, and when?"
Reference filtering returns the linked records, and issue activity comes back fully attributed, with an author and a timestamp on every entry.
Look across your runs.
"Which runs are paused right now?"
"Which runs started on June 1?"
Run-state and single-day date filters resolve cleanly: you get the count and the list behind it.
Open an assembly's bill of materials.
"What are the components of this assembly, with quantities?"
The full BOM comes back, each component with its quantity.
Tips to get the most out of it
A few things we learned that make the experience noticeably better:
Ask in plain language, then refine. You do not need to name a tool or a field. Ask the way you would ask a teammate, then follow up.
Search first, then open. The strongest pattern is to find a record by a human detail you know (a name, a code, a run number) and then ask the assistant to open it in full. It resolves the identifier for you, so you never have to memorize an internal ID.
Be specific about time. "Started on June 1," "paused right now." A precise window gives you a precise answer.
Ask for the whole list. When there are hundreds of matches, ask for all of them. A list request returns the complete set rather than a truncated first page.
Pivot from one record to related ones. Find a run, then ask which issues are linked to it. The assistant follows the references for you.
Safe by design
Read-only is not an afterthought here.
Every tool exposed in this MCP server is read-only today. It can look up and summarize your data; it cannot change a procedure, edit a run, or modify a part. There is no write path yet.
Access is granted through a standard OAuth sign-in, the same identity you already use. The assistant is scoped to exactly what your account can see, checked on the server side against your real permissions on every request. It cannot reach anything you could not already open yourself in Epsilon3.
Your data stays yours
When you ask a question, the assistant pulls only the records it needs to answer it, under your own Epsilon3 credentials, never a bulk export of your workspace. It can reach exactly what you could already open yourself, and nothing more, so a teammate's restricted data stays restricted. This MCP Preview is strictly read-only.
Getting started
If you use an MCP-compatible AI assistant, such as Claude, you can connect Epsilon3 as a server, sign in, and start asking. No new application to install on top of Epsilon3, and nothing to configure inside your procedures.
The MCP URL is https://mcp.epsilon3.io/
Add this URL as a custom connector to your Claude or Codex desktop or mobile app, and hit Connect.
This Preview is intentionally focused on reading: search, open, query, and trace. We are excited to ship it as a first step, gather your feedback, and use it to shape what comes next. The next planned version adds write tools, so the assistant can act and not just answer: sign off procedure and run steps, edit procedures, modify parts and inventory, and more. It is the foundation for a larger idea, which is that the system of record for how your hardware gets built should be as easy to ask, and eventually to update, as it is to fill out.
The data already runs your program. Now you can just ask it.
Frequently Asked Questions (FAQ)
-
Yes, in this Preview. It can only search, open, query, and trace your data; it cannot create, edit, or delete anything in Epsilon3. Write tools (signing off steps, editing procedures, updating parts and inventory) may come in the future only after thorough testing and guardrails.
-
No. You connect the server to your assistant once, then ask questions in plain language.
-
You sign in with your existing Epsilon3 identity over OAuth, and the assistant is scoped to exactly the data your account can already see, enforced on the server side.
-
Procedures, runs, parts and inventory, and issues.
-
Any AI assistant that supports the Model Context Protocol.
-
The assistant retrieves only the records it needs to answer your question, under your own credentials and permissions, the same data you could open yourself in Epsilon3.