​

Troubleshooting

​

Start of conversation tips

Most “the assistant is not using the Nexudus tools” issues come from how the conversation starts, not from a server problem. A few habits make a big difference:

​

Mention Nexudus by name early

Models like Claude and ChatGPT have hundreds of connectors, tools, and built-in capabilities available. If your opening message is “how many bookings do we have this week?”, the model may try to answer from general knowledge or web search instead of calling the Nexudus tools. A more reliable opening:

“Using the Nexudus tools, list bookings at my main location for this week.”

The word “Nexudus” and a hint at the action (“list bookings”) nudge the model straight to the connector.

​

Be specific from the first message

The Nexudus MCP server exposes 200+ entities through generic tools. Vague prompts force the model to guess which entity you mean, which often results in a wrong tool call or a clarifying question. Spell out:

• The entity — bookings, members, products, invoices, visitors, resources, plans.
• The filter — at which location, in which date range, with which status.
• The action — list, count, create, update, delete.

Compare:

​

Confirm the connector is actually connected

Before troubleshooting deeper, double-check that:

• The Nexudus connector is Connected in your client’s settings.
• In ChatGPT specifically, the connector is toggled on in the current chat’s Developer mode tool menu — it is off by default in every new chat.
• The 8-hour session has not expired. If it has, the assistant should say so on the first tool call — reconnect from your client’s settings.

​

Tell the assistant to use the tools, not guess

If the assistant answers from general knowledge instead of calling a tool, you can correct it:

“Don’t answer from memory — call the Nexudus tools to look up the current value.”

This is a common nudge with ChatGPT in particular, which sometimes treats Nexudus questions as general knowledge.

​

Connection issues

​

My client cannot reach https://mcp.nexudus.com

• Check that you typed the URL exactly: https://mcp.nexudus.com — no trailing slash, no extra path.
• Check your internet connection.
• Check that a firewall or proxy is not blocking outbound HTTPS connections to mcp.nexudus.com.

​

The login window does not open

• Most MCP clients use your default browser to show the Nexudus login page. Make sure you have a default browser set.
• Pop-up blockers can prevent the window from opening. Disable them and retry.
• In Claude Code and VS Code, check the log output for the manual login URL — you can open it in any browser to complete sign-in.

​

”Authentication required” or “Unauthorized” mid-conversation

Your 8-hour session has expired. Reconnect from your client’s MCP settings — see Sessions and re-authentication for the exact steps per client.

​

Login redirects back to a Nexudus error page

• Your Nexudus credentials may have changed. Try signing in directly at spaces.nexudus.com first.
• Your account may have been deactivated — contact your Nexudus administrator.

​

Tool issues

​

The AI assistant says no Nexudus tools are available

• Confirm the connector shows as Connected in your client’s settings.
• In ChatGPT, confirm the Nexudus connector is toggled on in the current chat’s tools menu — it must be enabled per chat.
• Restart your MCP client. Some clients only load tools at app startup.
• Disconnect and reconnect the connector.

​

The assistant picks the wrong field name or filter

The assistant should call nexudus_describe_entity before creating or filtering, but sometimes it skips that step. Ask it to describe the entity explicitly:

“Before updating that, describe the ‘products’ entity and list its filter keys.”

This pulls in the correct field names so the next call works.

​

List queries return wrong results

The Nexudus API uses specific filter keys (for example, Booking_FromTime rather than FromTime). The MCP server validates filter keys and returns an error listing the valid ones — the assistant should self-correct on the next attempt. If results still look wrong, ask the assistant to describe the entity and rerun the query using the listed filter keys.

​

A create or update fails with a validation error

• Ask the assistant to call nexudus_describe_entity and check which fields are required for the entity.
• Some entities have additional rules (price minimums, contract overlaps, location-specific defaults) that the API enforces. The error message usually says which field caused the problem.

​

Delete is too aggressive

If the assistant has deleted more than you intended, there is no automatic undo. Recovery options depend on the entity:

• For most entities, you would need to recreate them from a backup or from another source.
• For destructive bulk actions, always ask the assistant to list first so you can confirm the scope before deletion.

​

PII redaction issues

​

Names and emails come back as «PII:NAME:...» tokens

PII redaction is on for your session. This is the safer default. If you actually need to see real personal data, ask the assistant to switch the setting:

“Switch off PII redaction.”

The assistant calls nexudus_reauthenticate to revoke the session, and from there the flow is client-specific:

• Claude.ai / Claude Desktop — Claude shows a Connect button on the next pending tool call. Click it, sign in again, and tick the PII checkbox on the sign-in page.
• ChatGPT — ChatGPT prints manual steps: open Settings → Connectors → Nexudus, disconnect, reconnect, tick the PII checkbox on the sign-in page, then reply “done” in the chat.

See PII redaction → Switching mid-conversation for screenshots.

​

I want to turn PII redaction back on

Same flow as above — ask the assistant to switch redaction on. On the next sign-in, leave the “Allow the AI assistant to see member personal data” checkbox unticked.

​

Some names still appear even though redaction is on

Redaction covers data Nexudus returns. It does not sanitise:

• Text you type into the chat yourself.
• Names the assistant infers from your prompts (e.g. you say “Sarah’s booking”).
• Free-text fields that look like notes but contain non-PII content.

If you need full data minimisation, leave redaction on and avoid typing personal details into the prompt.

​

Performance issues

​

Calls feel slow

• The first call after connecting can take a moment as the session initialises and entity definitions are loaded.
• Long list calls take more time on large datasets. Ask the assistant to use pagination — “show the first 25 records” — and continue from there.

​

”Session not found” errors after a long break

• The server may have restarted while your session was idle. Reconnect from your client’s MCP settings — re-authenticating creates a fresh session.

​

Where to get more help

• Nexudus help centre: help.nexudus.com
• MCP server repository: github.com/Nexudus-Ltd/nexudus-coworking-mcp-stateful
• Entity reference: CLI entity reference — same entities are available through the MCP server.

​

Next steps