How do I verify Montage workspace access with Docker Agent?
Docker Agent can connect directly to remote MCP servers through YAML toolsets. That feature is distinct from Docker MCP Toolkit's catalog and gateway, so Montage does not need a fabricated container image or catalog listing. Its remote documentation specifies Streamable HTTP configuration and OAuth behavior, including differences between interactive use and API server mode. Start with a local attended agent session, not an unattended deployment.
Using Docker Agent, call only Montage signals_status as the first live operation. Report the actual workspace identity and coverage from the tool response. If authentication, transport, or the tool is unavailable, stop and explain the blocker. Do not start collection, recurring tracking, writes, publishing, messaging, or social actions.Choose direct remote MCP configuration
Inspect the Docker Agent YAML for the agent you will actually run. Under that agent's toolsets, use type mcp with a remote block. Do not add an unrelated server from Docker's catalog to represent Montage. The Toolkit's image signing and sandbox guarantees must not be transferred to a directly connected hosted service merely because both features use MCP.
Set the documented transport spelling
Inside remote, set url to https://app.montages.co/mcp and transport_type to streamable. Docker Agent uses that spelling in its current remote examples. Preserve existing toolsets and keep the endpoint application-controlled. A hosted connection does not launch a local Montage container, and changing container resource limits would not fix a failed remote authorization exchange.
Complete attended OAuth appropriately
Use the documented interactive OAuth flow for a compatible server. For servers without Dynamic Client Registration, Docker Agent documents additional configuration or a credential prompt; do not invent those values. API server mode delegates OAuth interaction differently and is outside this attended setup. Keep credentials in the existing secret mechanism rather than committing an oauth clientSecret in YAML.
Inspect the toolset without expanding access
Review the remote tool catalog and the agent's permission controls before allowing calls. Header interpolation and forwarded caller headers are advanced identity mechanisms, not shortcuts to workspace authorization. Do not forward arbitrary incoming credentials to Montage. A loaded toolset or cached OAuth token must still be checked against the intended user and the actual tool's permission requirements.
Request one status result
Make the first live Montage operation signals_status and inspect the returned workspace. If a cached token is rejected, use the documented refresh or reauthentication behavior without enabling broader approvals. Stop after the read-only check; automatic reconnection does not authorize repeating mutations. Collection, schedules, publishing, and social actions require separate explicit approval beyond successful remote MCP initialization.
Keep the boundaries clear.
- Direct remote toolsets are distinct from MCP Toolkit catalog/gateway; attended OAuth differs from API server mode.
- Documented setup, not live interoperability certification. A visible tool catalog is not workspace authorization.
- First live operation: read-only signals_status. No automatic collection, recurring tracking, or social actions; broader actions need separate explicit authorization.
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.