When it does not work

The failures below are the ones support sees most. Each entry gives the symptom first, because the symptom is what you have.

“Bad Data” or a cryptography error when you register

Symptom. Registration fails in Epicor with CryptographicException: Bad Data, and nothing in the message points at a key.

Cause. The signing key generated in Epicor came out with a wrong-length value, so it could never sign anything. This was a defect in the Epicor-side key generation. It is fixed, so a key created now is valid. You only see this on a key that was created before the fix.

Recovery for a key made before the fix

Re-registering reuses the existing key, so it keeps the bad one. To get a valid key, rotate the key or delete the stored key pair in Epicor and register again. With the generation fixed, the new key is valid. Rotating signs out everyone currently connected, so do it in a maintenance window. Contact us if you are unsure which rows to remove.

“Invalid RSA key format. Expected INTEGER.”

Symptom. Sign-in fails with that exact message.

Cause. The key stored in Epicor is in a shape the reader does not accept. This usually means an old key is still there from a previous setup.

Fix. Delete both stored key rows: the private and the public one: then register the instance again. Deleting only the private row leaves a stale public key behind. Ask us if you are unsure which rows to remove.

The assistant connects but has no tools

Symptom. The client shows the server as connected, but the assistant says it has no tools, or every request fails.

Check these in order:

1. Complete Connect, Sign in, or Authorize in the client’s connector settings.
2. Confirm the MCP URL points to the intended Epicor instance.
3. Ask the administrator to confirm the required pack tools or custom tools are installed and enabled.
4. Refresh the client’s tool list or reconnect. Make Cutova available in the current conversation.

For an access-denied error, check the user’s Epicor permissions and the connection’s API key scope. A connected server does not mean every tool or operation is available to that user.

A pack does not finish installing

• Solution not published: the pack’s Epicor download is not yet available. Contact support for availability.
• Missing pieces: install the pack solution in the selected Epicor environment, then run Verify pieces again.
• Tool capacity: review the tools that were not enabled and your plan’s available capacity.
• Other import failure: fix the reported cause and use Retry.

See Install a pack for the complete sequence.

A form or chart does not appear

Check that the built-in tool is enabled and that your client supports its display format. A client without MCP Apps support may receive text instead of an interactive panel. If a question hangs, check the client’s elicitation support and version. See Built-in tools.

“Session not found. Please reconnect.”

Symptom. The assistant worked, then started returning this.

Cause. Your session expired, or the server restarted.

Fix. The client should reconnect by itself. If it does not, disconnect and connect again in the client. You do not need to re-register anything.

Sign-in opens, then fails, in Copilot Studio

Symptom. The Epicor sign-in page opens, but authorization never completes.

Cause. Copilot Studio does not send the resource parameter, so ordinary discovery does not work.

Fix. Use OAuth type Dynamic with the path-based Authorization URL: the one with the quick code in the path. See Connect your AI client.

Everyone was signed out at once

Symptom. Every connected person stopped working at the same moment.

Cause. Almost always the instance was registered again with an older MCP library, which rotated the signing key instead of reusing it. Every token minted in the previous 90 days became invalid.

Fix. Everyone reconnects and signs in again. To prevent it, do not register an instance without the current library. See Register your Epicor instance.

“We do not recognise that code”

Symptom. You typed your quick code on the Cutova sign-in page and it came back with this message.

Cause. One of three, and the page cannot tell you which, on purpose.

• A typo. The code never contains I, L, O, 0 or 1. Those five are left out because they are easy to confuse, so if your code appears to hold one, you have misread a similar-looking character. Check it against the console.
• The wrong code, from a different organization or a different Epicor environment.
• The instance was removed from Cutova.

Fix. Check the code on the instance page in the console, or ask whoever set up Cutova for your organization. Case does not matter, and neither do spaces you did not type.

“This organization’s access is suspended”

Symptom. The Epicor pages of the Cutova console fail with this message. Signing in to the console still works, and the rest of it loads normally.

Your AI client shows something different. It cannot connect at all, and reports an authentication or authorization failure rather than this sentence. Reconnecting and signing in again does not fix it. If your console shows the message above and your AI client stopped working at the same time, they are the same cause.

Cause. Your organization’s access has been withdrawn by Cutova. This is deliberate, not a fault. Console sign-in keeps working so you can read why and reach someone about it.

Fix. Contact your Cutova administrator. Reconnecting the client, re-registering the instance and signing in again will not change it: the refusal is on the organization, not on your account or your tokens. Once access is restored, existing connections start working again with no action from you.

A colleague sees data that you cannot

Compare the selected company, plant, date range, and tool inputs first. User tool calls use the signed-in person’s Epicor identity, so different Epicor permissions can produce different results. Ask your Epicor administrator to review the applicable permissions when an expected record is missing.

Nothing here matches

Tell us:

• Which client, and which step.
• The exact message, with any error id.
• Your instance quick code: the four characters in your MCP URL.

Please do not send us a token or a password. We never need one.