A month ago we released the native connection with Claude via MCP: Hapee agents stopped calling Claude through the API and started invoking it as direct context. Lower latency, real multi-step reasoning, better use of customer context. That foundation is what makes it possible today to keep adding capabilities without rewriting anything.
This week Anthropic introduced Fable Mythos, a new variant of its Fable family. It's very recent — it has only been available for days — and we're still in the process of testing it thoroughly. But the direction is clear: it will join the stack of Fable models we already use in Hapee to boost agent reasoning.
What the Fable family is and how we use it
Hapee already integrates several Fable models as part of its reasoning layer. Each customer can choose the one that best fits their use case, and agents can switch between them depending on the task: quick classification, long conversation, structured reasoning, context retrieval from the CRM, reading the knowledge base, and so on.
The advantage of having them all behind the same Hapee MCP server is that switching between models is invisible to the end customer and, for the team operating the agent, it's a toggle in the settings. There's no need to rewrite prompts or move data.
What Mythos adds to the stack
Mythos is shaping up as a variant oriented toward continuity and extended reasoning in long conversations. In practice, that means agents that keep the thread better when the customer changes topic, returns to an earlier conversation, or enters a multi-turn funnel.
We're not promising specific benchmarks: the model is very new. What we can say is that Hapee's MCP architecture is designed so that, as soon as Mythos is production-ready, existing agents can adopt it without complex migrations. It's the same connection, with one more model.
Hapee doesn't depend on a single model. The native connection via MCP lets us incorporate each new Claude variant — including Fable Mythos — as soon as Anthropic releases it, without touching the logic of the agents already running in production.
The role of Claude Code along the way
Internally, the Hapee team uses Claude Code to iterate on the MCP server, maintain integrations, write documentation and test new variants as soon as they come out. It's part of what lets us be ready to incorporate new models in a matter of days, not weeks.
This isn't theory: it's the pattern we used to build the original MCP connection, the one we're using now with Mythos, and the one we'll keep using with every variant that comes along.
The Fable family + Hapee MCP · a flexible foundation for incorporating each new variant
What changes for Hapee customers today
In the short term: nothing you need to do. If you already run your Hapee agent with the native MCP connection, you're on the same infrastructure that will receive Mythos when it's ready for general use. We'll announce it on this same blog and in the product's version changelog.
For customers who haven't yet migrated to the MCP connection, the path is still the same: enable MCP in your workspace, connect context (CRM, conversations, knowledge base, calendar) and let the agents reason with the full information. Every Fable model we add to the stack runs on that same foundation.
Turn on Hapee's MCP connection in your account.
It's the foundation on which we add each new Claude variant. We'll show you in 25 minutes how it works, with your own CRM connected.
Book a demo →
