How do I verify Montage workspace access with Google ADK?
Google ADK uses McpToolset to adapt remote server tools into agent tools. The current documentation covers several languages, but its Python connection examples and deployment rules are specific enough to deserve a dedicated path. For Montage, define a stable remote toolset, filter its initial tools, and keep model credentials separate from the authorization that permits account-scoped workspace execution.
Using Google ADK, 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.Select the Python ADK interface
Review the installed google-adk release and the MCP extra documented for Python. Do not assume every TypeScript, Java, Go, or Kotlin example has identical constructor spelling. Identify the existing agent module and how its toolsets are closed. The documentation requires production toolsets to be defined synchronously in agent.py rather than relying on dynamic asynchronous initialization intended for local debugging.
Use the remote connection parameters
Create McpToolset with the documented Streamable HTTP connection parameters targeting https://app.montages.co/mcp. Keep StdioConnectionParams and npm launcher examples out of this remote definition. The toolset is an adapter between ADK and the server, not an alternative Montage service endpoint. Inspect the installed class names because older examples may use different capitalization or import paths.
Configure remote authentication deliberately
Review ADK's Remote MCP Authentication section and choose a mechanism compatible with Montage. Supply only existing approved credentials or a completed supported OAuth integration, not the model provider's API key. If the selected runtime cannot perform the required consent, stop before attaching usable tools. A deployed agent cannot assume an interactive browser is available just because local debugging had one.
Filter the discovered tools
Use the documented tool filtering facility to keep signals_status as the initial callable operation. Inspect its actual schema and any naming changes introduced by the adapter. A toolset that constructs successfully or returns definitions has not proved workspace authorization. Avoid attaching server resources or unrelated tools simply because the framework supports them.
Verify before the agent’s research role
Make the first live Montage operation signals_status and check its returned workspace against the configured identity. Close the toolset through the application's established cleanup path when testing ends. Keep failures explicit and do not substitute a model-only answer. Expanding an ADK agent to collection, recurring tracking, or social actions requires a separate approval boundary beyond this read-only connection check.
Keep the boundaries clear.
- This guide targets Python McpToolset; other ADK languages and older releases use different APIs.
- 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.