FR -- Add simulate_conversation functionality to the MCP server

[corbin-caldwell]

Simulate-conversation functionality would improve automation use cases for simulating calls, especially in the following scenario where our clients would need to automate analysis and performance of prompts/agents:

1. Get Prompt of agent-to-be-tested
2. Analyse Prompt, determine different possible user paths, create test-scenarios accordingly
3. Run these test scenarios (currently as http-request)
4. Analyze results and give suggestions

[corbin-caldwell]

The request in detail:

Claude could have an intelligent and active role in testing and improving a prompt [compared to manual n8n workflows]. But this only works with MCP.

"Imagine you would have a test feature for each agent that would automatically do the steps described above – so you could very efficiently test each change in the prompt.
Two options would be useful: 1) start from prompt, create necessary scenarios, 2) start from specific scenario defined by user.

If you are interested in feedback and input on product improvements, let me know.
Here some things that would tremendously help us with building agents at scale:

Global guardrails:
a bit like a knowledgebase, but as deterministic part of the prompt
we iterate our guardrails constantly to keep LLM in check, at the moment we have to update each new guardrail version in each prompt manually

Prompt Versioning:
We are working and improving an agent continuously, often with multiple people. Easy to lose track of prompt versions
Simulate_Conversation in voice mode:
the agents-to-be-tested seem to behave quite differently when tested with chat than they do with audio. We are observing utterances in chat mode we never get in voice mode and vice versa."