ontologies for AI & ML
MCP
Every major data platform is racing to become the control layer for AI agents. But protocol is not meaning.
Agents connecting to raw schemas still guess at joins, interpret metrics inconsistently, and operate without business rules or governance. Timbr’s MCP server exposes your ontology-based semantic layer instead, giving any MCP-compatible agent governed access to business concepts, semantic relationships, and measures across your entire data estate.
Timbr MCP Server
The Timbr MCP server connects any MCP-compatible AI client to your SQL ontology. Agents invoke tools like query_data, ask_question, and generate_sql over business concepts, relationships, and measures, with access controls and business logic enforced at query time. Authentication via API key or OAuth 2.0.
Timbr Helps Solve Challenges in MCP + Agent Data Access
| Challenge | How Timbr Helps |
|---|---|
| Agents hallucinate SQL over raw schemas | Timbr exposes business concepts not tables giving agents a governed semantic model to query against |
| Platform MCP servers are locked to one warehouse | Timbr's virtual ontology spans Snowflake Databricks BigQuery and other sources in a single unified model |
| Business logic lives in prompts or application code | Logic is defined once in the ontology and enforced at query time across every agent interaction |
| Governance is bypassed in AI-generated queries | Access controls row-level security and business rules are applied through the ontology automatically |
| Agents cannot explain how they reached an answer | Every query generates traceable SQL that maps back to named concepts relationships and rules in the ontology |
Ontologies + MCP: A Powerful Combination
When MCP meets Timbr’s SQL ontology, agent data access becomes:
- Governed: Agents query named business concepts with access policies enforced at the ontology level, not left to the agent to infer.
- Cross-platform: One ontology model spans multiple data sources – Snowflake, Databricks, BigQuery, and others. so agents can join data across platforms without moving it. No pipeline, no replication, no single-warehouse lock-in.
- Explainable: Every response traces back to concepts, relationships, and SQL defined in the ontology, creating a transparent chain of evidence.
- Reusable: Business definitions are modeled once and consumed by every agent, application, and workflow that connects through MCP.
- Secure: Row-level security, masking, and audit trails apply even to AI-generated queries issued through the MCP interface.
Unlike platform-native MCP servers, the Timbr MCP server works across your entire data estate and exposes the full depth of an ontology model, not just metric views. Agents get a semantic model that spans platforms, not just one warehouse ecosystem.
Why MCP Needs Timbr
| Without Timbr | With Timbr MCP |
|---|---|
| Agents query raw schemas and guess at joins | Agents query governed business concepts with explicit relationships |
| Business logic lives in prompts or application code | Logic is defined once in the ontology and reused by every agent |
| MCP access is limited to one platform's data | Ontology spans multiple sources in a single unified model |
| No governance on AI-generated queries | Access controls and business rules enforced automatically |
Use Cases
Natural Language Queries Across Enterprise Data
Connect any MCP-compatible agent to Timbr and let it answer business questions over Snowflake, Databricks, ServiceNow, and other sources through a single governed interface.
Governed Agentic Workflows
Deploy autonomous agents that operate within defined business boundaries, querying only the concepts, relationships, and data they are authorized to access.
Cross-Platform Semantic Grounding
Give agents a unified semantic model that works across your entire data stack, so decisions are not siloed inside a single warehouse’s MCP endpoint.
AI Copilots with Auditable SQL
Build LLM-powered copilots that generate and execute governed SQL through Timbr’s ontology, with every answer traceable back to defined business logic.
