The Enterpret MCP, upgraded: citations, Codex support, token efficiency
If you've ever asked Claude or Cursor a question about your customers and wondered where exactly that answer came from, this one's for you.
We just shipped a major upgrade to Enterpret's MCP server. The MCP server is our window for any agent to act on top of your customer intelligence, and as agents become how product teams actually work, the bar for what good looks like has gone up. Every answer needs to be evidence-grounded, every host needs to be supported, every session needs to hold up at depth.
Three things shipped. Here's why each one matters.
1. Every answer now has a citation, and real customer quotes behind it
Click any number an agent returns and you land in the exact dashboard view with the records that drove it. Ask for the voice behind a finding and the agent pulls the verbatim customer quote, with speaker attribution.
This is the differentiator. No other MCP server in this space sits on a real customer context graph, and that's what makes citations and verbatim quotes possible in the first place. Every other tool in the market dumps customer context into a model and hopes the answer is grounded. There's no way to verify anything.
We took the other path. Rather than using Claude on top of generic context, you get deep-link citations back to the evidence behind every number an agent returns. The difference between "5 users complained about onboarding" and "here's exactly what 5 of them said, in their own words" is the difference between a guess and something you can take into an exec review.
2. Now works with Codex, and a wider set of agents
The old setup required a /wisdom:initialize_wisdom ritual at the start of every session, plus usage guidelines loaded into the host. That preamble is gone. All the guidance the agent needs now lives inside the tool descriptions themselves.
The bigger story: removing that requirement unlocks hosts that didn't support it. Codex. A whole new set of agents your team can now collaborate with. Connect once, ask your question. That's it.
3. More queries per session, sharper answers, lower cost
Older versions could only get you two or three queries deep before the context window filled up. We restructured the schema and tool surface so the agent picks the right tool, pulls only the slices it needs, and writes sharper queries on the first try.
The result: smaller prompts, fewer hallucinated field names, and sessions that hold up at the depth real work requires.
If you've used Enterpret MCP before, the only thing that changes for you is this: stop running /initialize_wisdom, and start clicking the citations. Your existing chats and integrations keep working. Old tool names continue to resolve, so nothing in flight breaks.
