Model Context Protocol (MCP) is an open standard that allows an AI assistant to connect to external systems through purpose-built tools. After Epsilon3 is connected, you can ask a question as you would ask a teammate. The assistant determines which Epsilon3 read tool to use, retrieves the relevant records, and returns the result in the conversation.

The Preview supports six read-only capabilities:

You do not need to know an Epsilon3 API, select a tool, or provide internal record IDs. Start with the details you know—such as a procedure name, run number, part name, date, state, or teammate—and refine the request in follow-up questions.

Confirm the following before adding the server to an assistant:

Use this server URL when creating the connection: https://mcp.epsilon3.io/

In Claude, Epsilon3 is added as a custom connector. The exact navigation may vary by Claude application version, but the connection uses the same server URL and Epsilon3 authorization flow.

Add Epsilon3 as a custom connector in Claude using the Epsilon3 MCP server URL.

If you are not already signed in to Epsilon3, the authorization page opens. Enter your Epsilon3 email and password, or continue through your organization’s configured sign-in flow, then select Authorize. Selecting Deny cancels the connection.

Authorize the assistant to access Epsilon3 using your existing Epsilon3 identity.

After authorization completes, return to Claude and confirm that the Epsilon3 connector is enabled. You can then begin a new conversation and ask questions about the data available to your account.

In Codex, add Epsilon3 as a custom MCP server using Streamable HTTP. No bearer token, custom header, or environment variable is required because sign-in is completed through Epsilon3 OAuth.

Configure the Epsilon3 server in Codex as a Streamable HTTP MCP connection.

When prompted, complete the same Epsilon3 authorization flow shown above. Once authorized, Epsilon3 is available to Codex within the permissions of the signed-in account.

After the connection is active, ask questions in plain language. The assistant can search first, open records in full, and continue from one record to related records without requiring you to copy internal identifiers between requests.

Search for runs by participant, date range, exact start date, or current state.

Requests using “my” are matched to the Epsilon3 identity used during authorization.

“What were my runs in the last 7 days?”

“What ultimate test runs did Adam run within the last 2 months?”

“Which runs are paused right now?”

“Which runs started on June 1?”

The response can include both the number of matching runs and the complete list. For large result sets, explicitly ask for all matching records rather than a summary or sample.

A procedure can be found using a human-readable name or code and then opened in full. This includes its sections, steps, and telemetry checks, including large procedures that would otherwise be cumbersome to review screen by screen.

“Find the QA acceptance procedure and show me its full steps.”

“Compare the two takeoff procedures and show me where they differ.”

“Which steps in this procedure verify a telemetry parameter, and what limit does each one check?”

Issues can be filtered by status, opened in detail, and traced to the records they reference. Issue activity includes the author and timestamp for each entry, allowing you to understand who contributed and when.

“Show me the open issues, then the resolved ones.”

“List every open issue, all of them.”

“Which issues are linked to this run?”

“Who commented on this issue, and when?”

Search for a part by name, Epsilon3 ID, or tracking number, then ask for its available inventory.

For assemblies, you can also retrieve the full bill of materials and the quantity of each component.

“Check inventory for this part by tracking number.”

“What are the components of this assembly, with quantities?”

Once a record is in context, continue the conversation without starting over. For example, find a run, ask which issues are linked to it, open one of those issues, and then ask who commented on it. The server resolves direct relationships among parts, procedures, runs, and issues so you can follow the operational context across Epsilon3.

The most reliable workflow is to begin with the identifying information you know, review the search result, and then ask the assistant to open or analyze the exact record. You can continue refining the same conversation as additional context becomes available.

Use plain language, then refine. You do not need to name an MCP tool or database field. Ask the initial question naturally, then narrow the result with follow-up questions.

Search first, then open. Find a record by name, code, run number, participant, or tracking number before asking for its full contents.

Use precise time language. A specific date or window—such as “started on June 1” or “in the last 7 days”—produces a more precise result.

State when you need every result. For a large set, ask for the whole list or all matching records.

Pivot from a record to its relationships. After locating a run, procedure, part, or issue, ask for directly linked records instead of running a separate search.

Authorization uses your existing Epsilon3 identity through OAuth. Your Epsilon3 password is entered on the Epsilon3 authorization page and is not provided to the AI assistant. Each MCP request is checked against the permissions of the signed-in account, so the assistant cannot retrieve a record that the same user could not open in Epsilon3.

The assistant requests the records needed to answer the current question rather than receiving a bulk export of the workspace. Restricted information remains restricted according to the user’s Epsilon3 access.

The connection is managed in the AI assistant where the MCP server was added. Disable or remove the Epsilon3 connector or MCP server there when it should no longer be available. If the authorization session expires, reconnect the server and complete the Epsilon3 sign-in flow again.

The Epsilon3 authorization session may have expired. Reconnect or reauthenticate the Epsilon3 MCP server in the assistant, then retry the request.

Confirm that the MCP Preview is enabled for the correct Epsilon3 workspace and that the signed-in user has access to the expected records. Ask “Who am I and which workspace am I in?” to verify the active identity and workspace before repeating the search with more specific terms.

Confirm that the server URL is https://mcp.epsilon3.io/ and that the Epsilon3 MCP Preview is enabled for the workspace. For Codex, verify that Streamable HTTP is selected and that bearer token and header fields are blank. If the problem continues, contact Epsilon3 Support.

Is the Epsilon3 MCP server read-only?

Yes. It can search, open, summarize, query, and trace Epsilon3 data, but it cannot create, edit, sign off, or delete records.

Do I need to write code or learn an API?

No. Add the server once, authorize it, and ask questions in plain language.

What can it search?

The MCP can work with procedures, runs, parts and inventory, and issues, including direct relationships among those records.

Which assistants can use it?

The server can be used with assistants that support Model Context Protocol. The setup examples in this article cover Claude and Codex.

Can the assistant see more than I can see in Epsilon3?

No. Results are limited to the records and workspaces available to the Epsilon3 account used during authorization.

Can I use “my runs” or “my issues”?

Yes. The server can identify the signed-in user and use that identity to resolve requests written in the first person.

Will write actions be available later?

Write tools are planned for a future release, subject to additional testing and safeguards.