How do I diagnose a Continue Montage tool missing outside Agent mode?
Continue’s normal chat and Agent mode expose different tool capabilities. Resolve the interaction mode before changing credentials or server files, preventing a mode limitation from being misdiagnosed as remote authorization failure and avoiding unnecessary duplicate Montage registrations in the project. Resolve mode availability before interpreting credentials. Continue’s Agent-only MCP surface means that ordinary chat can lack tools even when the remote definition is correct and a separately authorized Agent session works.
Help me diagnose a Continue Montage tool missing outside Agent mode. Check whether the affected conversation runs in Agent mode. Request signals_status in the intended Agent session after approved auth. No writes.Identify the interaction mode
Check whether the affected conversation runs in Agent mode. A chat without MCP tools cannot establish that Montage or its account credentials are broken. The mode comparison should describe the affected interaction directly instead of inferring server health from whether a chat can mention Montage in text.
Inspect definition loading
Review relevant .continue/mcpServers/ files and their installed-version format. Determine whether the remote entry loaded rather than creating another server to compensate for the wrong mode. A loaded definition should be distinguished from a file merely present on disk, especially when several project and user configurations exist.
Confirm remote transport
Check the documented streamable-http definition and hosted endpoint. A local stdio example cannot provide this remote connection by placing an HTTPS URL into command fields. Keep Continue’s documented streamable-http spelling rather than assuming a transport property copied from another assistant expresses the same supported definition.
Review auth after availability
Once Agent mode exposes the intended server, assess version-supported authorization. Keep absent tools, rejected configuration and failed account access as distinct diagnostic outcomes. The diagnostic should first resolve tool availability, then evaluate supported auth; repeated credential attempts cannot give normal chat an Agent-only capability.
Verify the Agent path
Request signals_status in the intended Agent session after approved auth. Record the mode-specific result without claiming that ordinary chat gained the same MCP capabilities. A successful Agent result should remain Agent-specific, preserving the limitation for the ordinary chat surface that prompted the original investigation.
Keep the boundaries clear.
- Keep Continue’s documented streamable-http spelling rather than assuming a transport property copied from another assistant expresses the same supported definition.
- A successful Agent result should remain Agent-specific, preserving the limitation for the ordinary chat surface that prompted the original investigation.
Start with a read-only check: ask your agent to report actual source coverage before collecting records or exporting an audience.
Primary references
Documentation reviewed 2026-11-15. Product capabilities, client availability and plans can change. Third-party marks identify their products, not an affiliation.