Your teams are already building with Claude, ChatGPT, and Gemini. Halo MCP connects those tools to the data foundation Halo already built, so the apps, agents, and analysis you want to develop run on the numbers you can trust. Hosted by Crisp or in your own cloud.
Other AI models don’t understand your data, but Halo does. MCP gives every tool the same governed way in.
Your AI
Claude
ChatGPT
Gemini
Your vibe-coded apps
⇄
Halo MCP Server
Governance · Halo Knox
Calculations
Semantic models
Data foundation
⇄
Your data
Retailer data
Internal data
Syndicated data
Any other source
Read and write, in both directions. Hosted by Crisp, or running in your own Databricks, Snowflake, or Google Cloud environment.
Imagination needs a foundation
Anyone can point AI to a spreadsheet. The answers only hold up when the data underneath is accurate.
All your data, one view
Retailer, internal, and syndicated data, already ingested, normalized, and harmonized. Your AI reaches all of it through one governed connection, with no exports and no copy-paste.
Semantic models that speak retail
Halo understands items, stores, categories, hierarchies, and restatements. Your AI asks in business terms, and gets the right context, every time.
Calculations done once, done right
Sell-through, share, distribution, out-of-stocks, and promo lift are defined in the platform. Claude and ChatGPT use the same definitions as every Halo Desk, so the numbers match everywhere.
Read and write access
Read: ask in plain language and get trusted answers. Write: let agents and teammates update plans, log decisions, and trigger workflows. Every action is governed under Halo Knox.
Hosted or in your cloud
Use Halo MCP with Halo hosted by Crisp, or run it inside your own Databricks, Snowflake, or Google Cloud environment. Your data stays in your perimeter, and each source keeps its Knox class.
Conversational AI
Ask in plain language. Get answers you can act on.
Every answer comes from the same definitions your Crisp Halo solution Desks use, with the source named.
Illustrative example
Which of our top 20 items lost distribution at our largest retailer last month, and what did it cost us?
Three items lost distribution across 214 stores. Estimated lost sales: $1.2M.
Item A: 96 stores, mostly in two regions
Item B: 71 stores after a planogram reset
Item C: 47 stores, now out of stock
Source: Halo semantic layer · distribution and sell-through definitions · governed by Halo Knox
Build
Vibe code it. Halo makes it enterprise ready.
Quick prototypes are great for ideas. They stall when the numbers can’t be trusted and can’t be repeated. Connect them to Halo through MCP, and the same idea gets the infrastructure it needs.
A prototype on an export
Numbers re-derived in the prompt
A different answer from run to run
No access controls or audit trail
Lives on one laptop
The same idea on Halo MCP
Numbers from governed semantic models
The same answer, every time
Halo Knox permissions, on every read and write
Enterprise-grade infrastructure in your cloud
Ideas
When we say Halo can run retail, we mean it
Teams have starting points, not limits.
A weekly business review in one prompt
Ask for the story of your week and get it, with the numbers, the drivers, and the next actions.
A promo planner, production ready
The planner your team vibe-coded last month, now on governed data with repeatable results.
An agent that watches your shelves
Spot out-of-stocks and distribution losses early, and open the task for the right person.
Common questions
Which AI tools work with Halo MCP?
Any MCP-compatible client, including Claude, ChatGPT, and Gemini.
Is it safe to allow write access?
Writes follow the same governance as reads. Halo Knox sets who can access each source, and every action is permissioned.
Do I need Halo in my own cloud?
No. Halo MCP works with Halo hosted by Crisp and with Halo installed in your own cloud.
What happens to the prototypes my teams already built?
Connect them to Halo through MCP. The idea stays yours, and the data underneath becomes Halo’s.