Skip to main content
There are three ways to connect your Snowflake data to agents. The right choice depends on your engineering capacity and the nature of your use cases. In all three cases, the principle is the same: expose your Snowflake-backed capabilities as MCP tools, and the agent will call them at runtime.

​
Summary

  • Path A (Snowflake-native MCP): Use Snowflake’s managed MCP server to expose data directly — predefined queries for deterministic tools, Cortex Analyst for ad-hoc analytics. No custom code, no separate infrastructure.
  • Path B (custom MCP servers): Your team builds MCP servers for business logic beyond SQL — proprietary models, multi-step workflows, external service calls. Full control.
  • Path C (Mindset AI professional services): Mindset AI’s team helps build and hand over MCP server implementations. Same architecture as Path B, with implementation support.
Most production deployments combine paths.

​
Path A: Snowflake-native MCP

The fastest path to production. Snowflake hosts the MCP server for you — no custom code, no separate infrastructure to deploy or maintain. Deterministic tools: predefined SQL queries. Run CREATE MCP SERVER in your Snowflake account, point it at specific queries, and they’re available as callable tools. RBAC and governance are handled by Snowflake. Authentication is managed through a dedicated integration partner that handles OAuth token lifecycle and secure credential storage — your Snowflake credentials are never stored in the Mindset AI platform. You define exactly what each tool returns — no ambiguity, fully auditable. This is hours of configuration, not weeks of development. Ad-hoc analytics: Cortex Analyst. For questions that can’t be answered by a fixed query set, Cortex Analyst translates natural language into SQL using a semantic model you define. Snowflake reports 90%+ accuracy on real-world use cases. The key pattern: define verified queries in your semantic model — curated question-and-SQL pairs that Cortex Analyst uses as a foundation when generating responses to similar questions. Exposed through the same managed MCP server — no custom endpoints required. Both capabilities are configured within Snowflake and served through a single managed MCP server. Mindset AI calls it like any other tool. Snowflake’s RBAC and audit logging apply automatically.
Access control on this path operates at the role level — all requests authenticate as the same Snowflake role. Row-level security applies to that role’s permissions, not to individual end users. If you need per-user data scoping, see Path B or the Access control: roles vs users section in the overview.
This path is analyst-friendly — someone who knows the data can define tools, approved queries, and semantic models without needing a developer. Example: A travel platform uses Snowflake-native MCP to expose cancellation trend data and fare availability. Deterministic tools power specific lookups (route availability, fare classes). Cortex Analyst handles ad-hoc analytics questions about cancellation patterns and demand trends. Agents use both to surface proactive upgrade recommendations to customers — all without a single custom endpoint.

​
Path B: Custom MCP servers

For teams with business logic beyond what SQL can express.
  1. Your engineers write services that query Snowflake and execute business logic — proprietary scoring models, multi-step workflows, calls to other internal services.
  2. They expose those services as MCP tools with defined parameters and response shapes.
  3. They host the MCP servers in their own infrastructure.
  4. Mindset AI calls those tools at runtime.
Full control over every query, every business rule, every scoping decision. Best for transactional use cases (booking, enrolment, workflow triggers) and complex multi-step logic that involves systems beyond Snowflake. Example: A healthcare SaaS platform wraps their Snowflake assessment data, Learning API, and Competency API as three separate MCP tools. An agent queries all three in sequence to find qualified mentor candidates, then calls a fourth tool to create a group assignment — all from a single conversation.

​
Path C: Mindset AI tech enablement services

For teams that need help getting to production.
  1. Mindset AI’s implementation team builds MCP servers on your behalf, working with your team to define tool logic, scoping rules, and access controls.
  2. You retain ownership of the business logic and approve all tool definitions.
  3. Implementations are handed over to your team to own and host. Same architecture as Path B, with implementation support.
Example: A workforce management platform has strong Snowflake data but a small engineering team. Mindset AI’s implementation team builds MCP servers that query their scheduling, credential, and compliance data, enabling agents to answer questions like “who’s qualified and available for overtime next week” without the customer writing any integration code.

​
Combining paths

Most production deployments use a combination. Snowflake-native MCP (Path A) handles well-defined analytical queries and deterministic lookups. Custom MCP servers (Path B) handle transactional operations and multi-step logic that involves external APIs alongside Snowflake. The agent doesn’t know the difference — it calls whichever tools match the user’s intent.