Temporal MCP

Temporal is a durable execution platform: workflows are written as ordinary code, and the service replays them from an event history so they survive crashes, restarts and waits that last days. An MCP server for Temporal sits between an AI client and a cluster, whether that is a local development server or a Temporal Cloud namespace, and turns cluster operations into tools an assistant can call. The exposed surface follows the workflow lifecycle. An agent can list executions in a namespace with filters, describe a single execution, read its full event history, start workflows and fetch their results, send signals and run queries against running executions, and cancel or terminate them, including across batches. Schedules are covered too, so recurring workflows can be inspected and paused. Connection settings handle both a plain host and namespace for local work and certificate or API key authentication for Temporal Cloud. Debugging is the obvious payoff. Reading a long event history by hand is tedious work, and asking why an execution stalled, then having the assistant pull the history and the pending activities itself, beats assembling CLI invocations. The same reach makes it a sharp instrument, since terminate and signal change production state and a batch operation multiplies that. Scope the credential you hand an agent to a specific namespace, and keep write capable connections away from clients you let run unattended.