How do I verify Montage workspace access with Haystack?
Haystack's MCPToolset integration comes from mcp-haystack and uses a server_info object to describe the connection. It can provide tools to a Haystack agent, but the transport class and authorization setup remain explicit application choices. This guide uses the documented StreamableHttpServerInfo path for Montage and avoids implying that the MCPToolset reference establishes a generic browser OAuth experience or a live-tested deployment.
Using Haystack, 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.Match the integration package to Haystack
Review the mcp-haystack integration and MCPToolset documentation against the application's existing Haystack version. The connection belongs in the tool integration rather than a document-store configuration. Identify the component that will own the toolset and decide how its lifetime ends. Do not install a similarly named third-party package or assume that core Haystack alone supplies every transport helper shown in the integration reference.
Construct the correct server_info
Use StreamableHttpServerInfo for https://app.montages.co/mcp and pass it as server_info to MCPToolset. The reference also lists SSEServerInfo and StdioServerInfo, which describe different connection types. Do not substitute one merely because a sample uses it. A local command-based server and a hosted Streamable HTTP service have different runtime requirements, and this integration does not require launching a Montage process in the pipeline.
Establish authentication before agent wiring
The MCPToolset page does not establish generic OAuth discovery, consent, and refresh for Montage. Check the installed server-info and transport interfaces and supply only a supported existing authorization mechanism. If no compatible flow or approved credential is available, leave the toolset blocked. Neither a Haystack agent configuration nor credentials used for its model provider authorize a private Montage workspace or justify forwarding unrelated identity headers.
Inspect the selected toolset
Review the actual tools loaded from the server and select the status operation for the first agent or direct-call test. Keep broader research tools out of the initial pipeline and preserve mutation approval at the application boundary. Discovery describes available schemas, not execution permission. Avoid treating a component's successful construction as connection proof, because initialization, authorization, and a later workspace tool call can fail independently.
Test status outside a research loop
Execute signals_status as the first live Montage operation and validate the returned workspace identity before passing results to subsequent components. Preserve transport and authorization errors explicitly rather than converting them into an empty document list. End the diagnostic run without collection, schedules, or social actions. A later agent loop needs its own bounded tool policy; this setup guide does not certify every Haystack deployment as interoperable.
Keep the boundaries clear.
- MCPToolset comes from mcp-haystack; the server-info reference does not establish generic interactive OAuth.
- 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.