Troubleshooting

Common issues and solutions when using the a4b.ai MCP server with your AI assistant.

Authentication Errors

Token expired or invalid

Your AI client shows an authentication error or stops responding to a4b.ai requests.

Solution: Most MCP clients refresh tokens automatically. If the error persists:

Remove and re-add the a4b.ai MCP server in your client
Re-authorize when prompted
Check if the application was revoked in Settings > Connected Apps

Insufficient permissions

Your AI client returns “Insufficient scope” when trying to create or update resources.

Solution: Grant mcp_write as well as mcp_read. Sign in again when your client next prompts you, and select both scopes on the authorization page.

A token carries the scopes that were approved when it was issued, so changing what an already-connected client can do means going through authorization again — re-authorize: sign in, approve access again, and select the scopes you want on the consent page.

Waiting will not do it. A connected client keeps renewing its access quietly in the background, and a renewal only ever returns the scopes it already had, so it never shows you a consent page again. You have to start a fresh authorization: revoke the application under Settings > Connected Apps, then reconnect it in your AI application.

“Insufficient scope. Required: mcp_destructive”

force_delete_asset returns this when called. Note the tool is listed for every client, whether or not the token can use it — so seeing it in your tool list does not mean you have access.

Solution: Re-authorize and select mcp_destructive on the consent page. It is never preselected, and the page will not accept your submission until you also tick the acknowledgement next to it.

What your client asked for does not limit this. The consent page offers every supported scope regardless of the scopes the client named at registration or in its authorization URL, so a client that only ever asks for mcp_read mcp_write — or names no scope at all — can still be granted permanent deletion.

Holding the scope is not sufficient on its own. It lets the call through the scope check; the asset still has to be one you may permanently delete as an administrator, which is checked separately on every call — see Permanent deletion is refused below. Permanent deletion also remains available in the web interface, on the asset’s edit page, under “Danger zone”.

Permanent deletion is refused

force_delete_asset returns an error even though your token has mcp_destructive.

Solution: Check each of these in turn.

Confirmation mismatch — the call must include confirm_inventory_number matching the asset’s inventory number exactly. Read it with get_asset first rather than guessing; a mismatch deletes nothing.
Not an administrator — restricted to organization administrators and administrators of the workspace containing the asset, the same as in the web interface.
Wrong organization or workspace — the asset must be one your token can reach.

If you meant to retire an asset rather than destroy it, use update_asset_state with state deleted instead. That keeps the asset and its history, and is reversed by setting the state back to available or in_use.

Not authorized for a resource

Your AI client returns “Not authorized” for a specific workspace or action.

Solution: Your user role doesn’t have permission for this action. Contact your organization admin to grant the appropriate role via Settings > Workspace Members in a4b.ai.

Data Errors

Resource not found

Your AI client can’t find an asset, workspace, or other resource.

Causes:

The resource ID doesn’t exist
The resource belongs to a different organization
The resource has been deleted

Solution: Ask your AI assistant to list resources first (e.g., “List my assets”) to find valid IDs.

Validation failed

Your AI client returns validation errors when creating or updating resources.

Solution: Check the required parameters and accepted formats for the tool. Ask your AI assistant “What parameters does [tool name] accept?” to see the tool’s schema.

Connection Issues

Client doesn’t show a4b.ai tools

Verify the server URL is exactly https://a4b.ai/mcp
Restart your MCP client after configuration changes
Check that your config file is valid JSON
Check client logs for connection errors
If the tools appear but a call fails with “Insufficient scope”, that is a scope issue rather than a connection one — see Insufficient permissions or, for permanent deletion, “Insufficient scope. Required: mcp_destructive”

Check that your browser isn’t blocking popups from the MCP client
Ensure you have an active a4b.ai account

Token refresh fails

Refresh tokens are single-use — each refresh returns a new one
If your client used an old refresh token, re-authenticate through the full OAuth flow
Check if the application was revoked in Settings > Connected Apps

Plan Limits & Authorization

Plan limit reached

Your AI client returns an error when creating assets, workspaces, or inviting users.

Solution: Your organization has reached the capacity limit for the current plan. Check your usage with “Show my organization stats” — if you’re at capacity, upgrade your plan or remove unused resources.

PKCE failure (`invalid_grant`)

Your MCP client fails during the OAuth flow with invalid_grant or similar errors.