Processes exposed as agent tools

With https://github.com/camunda/product-hub/issues/3353, process designers can expose processes as AI agent tools.

Primarily, an additional MCP endpoint in the Orchestration Cluster at /mcp/processes offers the exposed processes as MCP tools, so AI agents can consume them via MCP.

However, it could also be interesting for AI agents that already use c8ctl to consume those processes directly via c8ctl, instead of connecting to the MCP client additionally and authorizing against it (which could happen via c8ctl of course, but it's still another MCP connection for the agent).

To achieve this, c8ctl would need to offer the processes exposed as agent tools in some way. They can be found/listed in the Orchestration Cluster by searching for message subscriptions that a) are not in messageSubscriptionState DELETED, b) have messageSubscriptionType START_EVENT, and c) have a toolName.

The tool name is the name of process as tool. Further details on what is defined for exposed processes can be found in https://github.com/camunda/camunda/blob/main/gateways/gateway-mcp/src/main/java/io/camunda/gateway/mcp/processes/ProcessesToolRepository.java.

To be defined:

• Command under which agent processes can be found in c8ctl
• Concrete data and structure for agent processes in c8ctl

Start at the c8ctl command entry points and review ProcessesToolRepository.java for how exposed processes are identified and described. Define the command location and the concrete process data structure, then verify that the proposed c8ctl interface can represent the listed process tools and their details.