n8n "failed to refresh MCP registry": test Sume directly

A registry refresh error in n8n does not tell you whether your Sume MCP server works. Test Sume directly: connect to https://mcp.sume.com/mcp, call mcp_health, then call tools_list. If both succeed, the problem is the registry or n8n, not Sume.

Why you may see it now

The n8n releases page lists n8n 2.42.0 on September 29 with the note "Remove feature flag for MCP Registry in AI Assistant". The cause of the error message is not documented there, so this post makes no claim about it.

Separate registry errors from server errors

A registry lists servers; it does not run them. Two different failures can look alike in a workflow tool: the registry cannot be fetched, or a server you configured cannot be reached or authenticated.

• Registry failure: no server is listed or refreshed. Your own server entry may still work if you add it by URL.
• Server failure: the connection or tool call fails with an auth, scope or network error.
• Test the server without n8n in the loop, for example with a different MCP client, so you have a known-good baseline.

What to call on Sume

Sume's hosted MCP provides mcp_health for endpoint readiness, auth source and safety posture, tools_list for every tool visible in the session, and tools_schema for one tool's contract. Under OAuth with only mcp:read, mutating and paid tools are hidden, so a short list may mean a scope issue rather than a failure. API-key sessions see the full set.

If a tool call fails, read the error code: insufficient_scope means the session lacks mcp:write, and paid calls always need an idempotency_key.

Next step

If Sume answers correctly, configure the server in n8n by URL and credential rather than waiting for the registry, and report the registry error to n8n with your version number.