MCP servers: not a trigger source

A connected MCP server is a read-only tool source. It cannot open a Landfall incident, and this page exists to say so and why, rather than leave the question open.

Why there is nothing to configure

• The protocol has no outbound signal. An MCP server answers requests a client makes (tools/list, tools/call). It never delivers anything to a webhook on its own, so there is no delivery Landfall's trigger pipeline could verify, normalize and deduplicate.
• Landfall only calls read-only tools. Every operation it offers the investigator is one the server annotated readOnlyHint: true. A tool that would subscribe to or emit alerts is not read-only and is never called.
• The provider behind the server keeps its own path. Grafana's alerting, Sentry's issue alerts, Datadog's monitors: each opens a war room through that provider's own triggering integration, unchanged by whether you also reach it through an MCP server. See the other pages in this section.

What this means for the four-part integration requirement

Landfall holds every integration to four requirements (setup guide, dynamic agent access, incident triggering, and the shared architecture page). For MCP servers the third is met by recording it as not applicable with the reason above, not by inventing a trigger a server cannot send. The other three apply in full.