# Boreon™ · How the Boreon MCP Layer Works

URL: https://boreon.com/mcp/layer
Language: en

> The Boreon MCP Layer: one semantic layer over Tableau Server or Tableau Cloud, Databricks and Snowflake, with agentic chat, migration, fidelity and governance.

---
Boreon MCP Layer

# One semantic MCP layer over all three.

The layer connects to your Tableau Server or Tableau Cloud site, Databricks and Snowflake through their own APIs. It learns how your Tableau fields map to warehouse columns and keeps that map current. It only writes when you ask, and every write is logged.

[Proof and Fidelity](https://boreon.com/mcp/proof) [Book a working session](https://boreon.com/contact)

[Ask Tableau Anything](https://boreon.com/mcp/tableau)

## One layer over your whole Tableau environment.

People ask in natural language. The Boreon MCP Layer turns each question into reads against Tableau's own APIs and brings the answer back into the chat. The model underneath is yours to choose.

Your people

- Executives
- Analysts
- Site admins
- In six languages

Boreon MCP Layer

- Natural-language chat
- Answers in line
- Governance scans
- Site operations

Your LLM: Swiss-hosted, Claude, OpenAI or your endpoint

Tableau Server or Tableau Cloud

- REST API
- VizQL Data Service
- Metadata API
- TSM, on Tableau Server

Publish, the one write

## One layer that speaks every platform.

The layer reads each side through its own APIs and keeps a map of how Tableau fields line up with warehouse columns. It only writes when you ask, and it logs every write.

Tableau

- Tableau Server or Tableau Cloud, your own site
- Workbooks and dashboards (.twb / .twbx)
- Published data sources (.tds)
- Tableau Prep flows (.tfl)
- VizQL Data Service and the Metadata API
- PAT, JWT or a central service account

Boreon MCP Layer

Behind the Data & AI Portal

Databricks

- Your workspace and Unity Catalog
- Unity Catalog tables and SQL warehouses
- AI/BI dashboards, with Genie built in
- Materialized views from Prep flows
- Statement Execution API for every query
- PAT, OAuth or a service principal

Snowflake

- Your account, databases and roles
- Databases, schemas and virtual warehouses
- Streamlit in Snowflake apps
- Dynamic tables from Prep flows
- SQL API for every query
- Key pair, OAuth or a PAT

## Six engines behind one chat.

Migrate DBX Platform Latency Calls & Uptime to Databricks

1. ### Agentic chat
  Ask in natural language. You see what it is doing as it works, and answers come back as live tables and charts.
  Took the request in natural language
2. ### Semantic layer
  Tableau fields mapped to Databricks and Snowflake columns once, then kept current.
  Matched Tableau fields to Databricks columns
3. ### Lineage
  Traces every dashboard back through its data source and Prep flow to the column.
  Found the Prep flow behind the workbook
4. ### Migration engine
  Plans the move of each workbook, dashboard and Prep flow from Tableau Server or Tableau Cloud, then carries it out.
  Built the view first, then the dashboard
5. ### Fidelity engine
  Checks the warehouse's numbers against Tableau's, row by row, through the VizQL Data Service.
  41 of 41 rows match Tableau
6. ### Governance
  Read-only until you say otherwise. Your SSO roles carry through, and every write is logged.
  Every write logged under your SSO roles

[Your LLM, Your Rules](https://boreon.com/mcp/tableau-llm)

## Each task on its own model.

The platform routes four kinds of work separately. Keys are encrypted at rest, and every routing change is audited.

|  | **Claude** Anthropic | **GPT** OpenAI | **Any** Your endpoint |
| --- | --- | --- | --- |
| Chat | Chat · Anthropic Claude | Chat · OpenAI GPT | Chat · Your endpoint Any |
| Drafts | Drafts · Anthropic Claude | Drafts · OpenAI GPT | Drafts · Your endpoint Any |
| Escalation | Escalation · Anthropic Claude | Escalation · OpenAI GPT | Escalation · Your endpoint Any |
| Classify | Classify · Anthropic Claude | Classify · OpenAI GPT | Classify · Your endpoint Any |

Illustration · sample data

[The Portal That Stays](https://boreon.com/mcp/portal)

## For everyone from analysts to executives.

People who have never written SQL get live tables and charts by asking. And because the layer knows Tableau, Databricks and Snowflake, the portal comes with you when you move from Tableau Server to Tableau Cloud, and keeps answering from all three.

## People see what they are allowed to see.

SSO roles carry through to both warehouses, and the layer is read-only until you ask it to write. Every write is logged.

Signed in through your SSO

Analyst, Europe Executive, all regions

Which region had the most failed calls last week?

| Region | Failed calls |
| --- | --- |
| us-east | 142 Not in this role |
| eu-central | 97 Not in this role |
| ap-northeast | 63 Not in this role |
| us-west | 28 Not in this role |

Illustration · sample data

[Einstein Trust Layer](https://boreon.com/mcp/einstein-trust-layer)

## Five protections, on every call.

The Einstein Trust Layer sits between the Boreon MCP Layer and the model. Every call takes the same path, in the same order, every time.

What you see

How did **Maria Keller**'s territory do last quarter?

**Maria Keller**'s territory grew 12%, led by two renewals.

Einstein Trust Layer

1. Grounded
2. Masked
3. Zero data retention
4. Toxicity checked
5. Audited

What the model sees

How did **[PERSON_1]**'s territory do last quarter? + 14 records you may see

**[PERSON_1]**'s territory grew 12%, led by two renewals.

Once it answers, the model keeps none of it.

Illustration · sample data

1. 01
  Grounded in what you may see
  The prompt is grounded in your org's data, and only in the records the person asking is allowed to see.
2. 02
  Personal data masked
  Names, email addresses and other personal details become placeholders before the prompt reaches a model.
3. 03
  The model keeps nothing
  Zero data retention: the model provider stores none of the prompt and none of the answer.
4. 04
  Toxicity checked
  Every answer is scored for toxicity on its way back, before anyone reads it.
5. 05
  Audited in Data 360
  The whole exchange is written to an audit trail in Data 360, so it can be reviewed later.

[Agentforce Agent](https://boreon.com/mcp/agentforce-agent)

## When a question needs Tableau, Agentforce asks ours.

Your people talk to Agentforce inside Salesforce. When the answer lives in Tableau, the agent asks the Boreon MCP Layer, which reads your Tableau Server or Tableau Cloud site and does the Tableau Next work.

**Your people** In Salesforce, in natural language **Agentforce** Your agent, in your org **Boreon MCP Layer** Knows your Tableau

**Tableau** Server or Cloud, read through Tableau's own APIs **Tableau Next** Visualizations and dashboards, in Salesforce

[Where It Lives](https://boreon.com/mcp/architecture)

## Every piece stays on your side.

Outside your cloud

**Your people** Browser, on your network or VPN **Your SSO** Entra ID, Okta or any OIDC

Your Azure VNet or AWS VPC · private subnet

**Boreon MCP Server** Your central LLM portal

**WAF edge** Caddy + OWASP Coraza rules **MCP server** Chat, migrations, fidelity **Postgres** pgvector · Knowledge, audit, secrets

**Your OpenAI keys** Azure OpenAI or OpenAI **Your storage** S3 or Blob, for backups

Your platforms

**Tableau** Server or Cloud, over REST and VizQL **Databricks** Unity Catalog, over Private Link **Snowflake** Your warehouses, over Private Link

Hover or tap any part to trace what it talks to.

- ### WAF edge
  Caddy with OWASP Coraza rules in front of everything.
- ### MCP server
  Chat, migrations and fidelity checks in one set of containers.
- ### Postgres + pgvector
  Knowledge, the audit trail and secrets, behind your firewall.
- ### Your OpenAI keys
  Azure OpenAI or OpenAI, on your own keys and your own account.
- ### Your storage
  Backups land in your own S3 or Blob storage.
- ### Your SSO
  Entra ID, Okta or any OIDC provider. Roles carry through to both warehouses.

## The same layer, every step of the way.

The layer already knows Tableau, Databricks and Snowflake. When you move from Tableau Server to Tableau Cloud, it comes with you: the same questions, the same semantic model, and Databricks and Snowflake still feeding Tableau.

Today, on Tableau Server

After the move to Tableau Cloud

[See the portal](https://boreon.com/mcp/portal)

[Previous Overview Boreon MCP Layer](https://boreon.com/mcp) [Next Proof and Fidelity Boreon MCP Layer](https://boreon.com/mcp/proof)

## Which Tableau dashboard would you start with?

Start there. We'll put the layer over it and answer its questions in natural language, and if you ever want it on another platform too, every number is checked against Tableau first.

[Book a working session](https://boreon.com/contact) [See it live at ai.boreon.com](https://ai.boreon.com/)