For the past year, the question Hapee customers asked most was always the same: "why does the agent sometimes seem less intelligent when I give it a lot of customer information?". The technical answer was frustrating: AI agents on traditional platforms trim the context before sending it to the model, because every token costs money. The result: the model doesn't see the whole conversation, doesn't see the CRM, doesn't see the purchase history. It loses nuance.
The Model Context Protocol (MCP), which Anthropic published as an open standard, solves that. Instead of sending a fixed prompt, MCP lets Claude request additional context in real time from external sources (CRM, knowledge base, tools). Reasoning becomes truly multi-step.
Before: the agent calls the model once with a fixed prompt. Now: the model asks for the data it needs when it needs it, reasons, asks for more, decides. Lower latency, better answers.
What MCP Native is
MCP Native (native connection with Claude) is the implementation of Anthropic's protocol inside Hapee agents. When an agent is set to MCP mode, it automatically exposes the following available contexts to Claude:
- the CRM — contact, tags, history, pipeline stage, estimated value
- Conversations — the entire cross-channel conversation history for that contact
- the Knowledge Base — documents, FAQs and manuals uploaded to the workspace
- Calendars — your team's real availability, not simulated slots
- Workflows — the ability to trigger workflows from within the conversation
How it shows up in practice
A concrete example: a customer who has already bought 3 times writes "hi, I need help with my order". With a classic agent, the model receives that phrase and a fixed system prompt. With MCP, Claude first asks the CRM whether this contact has open orders, sees there's one "in transit", sees the estimated delivery date, sees a previous ticket the customer opened yesterday. And only then does it respond:
"Hi Camila! I see your order #4892 is leaving the warehouse today and should reach you tomorrow between 10:00 AM and 2:00 PM. Is that why you're writing, or is it something else about the ticket you opened yesterday regarding the size?" — Response generated with MCP Native
The difference is that it feels human. Because it has all the context, not a trimmed prompt.
Performance · real numbers
After 4 weeks in beta with 12 customers:
- Average latency: 1.4s (vs 2.8s with the traditional API)
- Cost per conversation: −22% (Claude only requests the context it needs)
- First-contact resolution: +34% (the agent closes more cases without escalating)
- Perceived quality (end-customer survey): +41% better
How to turn it on
Available to all Hapee customers starting today:
- Settings → AI Models In your agent's model selector, choose
Claude (MCP Native)instead of the traditional API options. - Connect the contexts By default Hapee exposes CRM, Conversations, Knowledge Base and Calendar. You can disable any of them individually if you don't want the model to access something.
- Test in the sandbox A test conversation with your own number. You'll notice the lower latency right away.
- Go to production Turn on MCP for all the agents in a workspace with a single toggle.
What's next
MCP Native is the first of three native integrations coming this quarter:
- OpenAI Assistants API native · tool-use functions with persistent state
- Gemini live · real-time multimodal voice sessions
- Mistral Codestral · agents that write workflow code from natural language
But MCP is the most important one. It's the first real industry standard, and Anthropic released it as open source so any platform can adopt it. That Hapee is the first in LATAM to have it natively isn't just marketing: it's infrastructure.
Turn on MCP Native in your account.
We'll schedule a 25-minute demo · we'll show it to you with your own CRM connected.
Book a demo →
