How do I verify Montage workspace access with OpenAI Responses API?
OpenAI Responses can use a hosted MCP tool directly, with server_url and server_label describing the remote service. This is not the OpenAI Agents SDK or a local Codex configuration. Montage authorization must be supplied through the documented remote tool mechanism, and approval requests remain part of the response lifecycle. A model response that merely names tools is not proof that a workspace operation executed.
Using OpenAI Responses API, 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 the hosted MCP tool path
Use the current Responses API MCP guide and an already configured API application. Do not create a second local MCP client solely to reproduce a provider-hosted tool. Confirm the selected model and API version support the intended MCP configuration, and review what data the remote service will receive before attaching it to a conversation.
Set a recognizable server definition
Configure type mcp with server_label identifying Montage and server_url set to https://app.montages.co/mcp. Keep the address trusted and stable. The hosted API must reach that endpoint itself; a URL reachable only from your laptop would not suffice. Do not claim that registration in an OpenAI request makes Montage a verified connector or published application.
Supply authorized remote access
The tool's documented authorization field accepts the remote credential needed by an authenticated server. Your application is responsible for obtaining and managing a compatible approved authorization, including OAuth lifecycle where required. Do not put an OpenAI API key into that field or expose remote credentials in user-visible output. Without an established Montage grant, leave execution blocked.
Constrain discovery and approval
Use allowed_tools to limit exposure to the exact signals_status name and keep approval required for the diagnostic. Inspect a real approval request before responding to it in the documented continuation flow. Neither a discovered MCP tool list nor an approval to send one request proves that all workspace tools are permitted.
Inspect the tool-call item
Make signals_status the first live Montage operation and examine the actual MCP result item for workspace identity and errors. Do not rely solely on response.output_text, which is a model summary. Continue no further on rejected authorization. This guide performs no paid request and authorizes no automatic collection, recurring research, publication, or social action after the status check.
Keep the boundaries clear.
- Provider-hosted MCP requires separate remote authorization and API access; this guide makes no paid request.
- 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.