Data 360, Tableau Next, and CRM Analytics MCP

Overview#

When enterprise AI agents answer questions like “What was our net retention rate in APAC last quarter?” or “Which high-value accounts have opened cases but zero website activity in 30 days?”, standard SObject queries (platform/sobject-reads) fall short. The relevant data often lives in federated data lakes, unified profile graphs, or multi-gigabyte time-series analytics datasets.

Salesforce provides three specialized Hosted MCP Servers for analytical workloads:

1. Data 360 Query Server (data/data-cloud-queries): Executes federated ANSI SQL queries across unified customer profiles and data lake tables.
2. Tableau Next MCP Server (analytics/tableau-next): Interacts with governed Semantic Data Models (SDMs) and runs natural-language Concierge Analytics queries without writing raw SQL.
3. CRM Analytics MCP Server (analytics/crma-beta, Beta): Inspects Wave dataset Extended Metadata (XMD) and executes high-speed Salesforce Analytics Query Language (SAQL) calculations.

Data 360 Query MCP Server (data/data-cloud-queries)#

Organizations using Salesforce Data 360 (formerly Data Cloud) can query Lakehouse objects, Data Model Objects (DMOs), Data Source Objects (DSOs), and Calculated Insights directly through the data/data-cloud-queries Hosted MCP Server.

Server URLs#

• Standard Global Gateway: https://api.salesforce.com/platform/mcp/v1.0/data/data-cloud-queries
• My Domain Route: https://<my-domain>/services/mcp/v1/d/data/data-cloud-queries

Prerequisites#

• Active Data 360 license provisioned in the org.
• Authenticated user or ECA Connected App assigned the Data Cloud User or Data Cloud Admin permission set, including the View Data 360 permission.

Tools Reference#

SELECT 
    UnifiedIndividual__dlm.ssot__FirstName__c,
    UnifiedIndividual__dlm.ssot__LastName__c,
    UnifiedContactPointEmail__dlm.ssot__EmailAddress__c,
    Customer_LTV_Insight__cio.Lifetime_Value__c
FROM UnifiedIndividual__dlm
LEFT JOIN UnifiedContactPointEmail__dlm 
    ON UnifiedIndividual__dlm.Id__c = UnifiedContactPointEmail__dlm.PartyId__c
LEFT JOIN Customer_LTV_Insight__cio 
    ON UnifiedIndividual__dlm.Id__c = Customer_LTV_Insight__cio.Individual_Id__c
WHERE Customer_LTV_Insight__cio.Lifetime_Value__c > 50000
ORDER BY Customer_LTV_Insight__cio.Lifetime_Value__c DESC
LIMIT 100

Tableau Next MCP Server (analytics/tableau-next)#

Direct SQL querying against raw tables introduces severe hallucination risks—LLMs frequently misinterpret complex revenue recognition rules or join foreign keys incorrectly. The Tableau Next MCP Server solves this by routing questions through governed Semantic Data Models (SDMs).

Through Tableau Next and Concierge Analytics, business definitions (such as ARR, churn rate, or gross margin) are certified once in the semantic model. The AI agent inspects certified metrics and dimensions before requesting analytical answers.

Server URLs#

• Standard Global Gateway: https://api.salesforce.com/platform/mcp/v1.0/analytics/tableau-next
• My Domain Route: https://<my-domain>/services/mcp/v1/d/analytics/tableau-next

All 15 Tableau Next Tools by Functional Group#

1. Analysis & Q&A Tools

• analyze_data: The primary conversational analytics tool. Accepts utterance (e.g., “Show revenue by product family for 2026”), target_entity_type (typically "sdm"), and target_entity_name_or_id. Executes natural-language Q&A backed by the governed semantic layer.
• list_dashboards: Lists all Tableau Next dashboards accessible to the authenticated user.
• get_dashboard: Retrieves metadata, layout, and component bindings for a dashboard by dashboardIdOrApiName.
• list_visualizations: Discovers available standalone visualizations and worksheets.
• get_visualization: Retrieves a specific visualization definition. Note the British spelling: The parameter name is visualisationIdOrApiName (with an s). Passing visualizationIdOrApiName with a z will throw a schema validation error.

2. Semantic Model Structure Discovery

• list_semantic_models: Returns all published Semantic Data Models (SDMs) in the workspace.
• get_semantic_model: Retrieves full schema details for a model using modelIdOrApiName.
• list_semantic_model_data_objects: Lists physical and logical tables mapped within an SDM.
• list_semantic_model_relationships: Inspects joins, cardinalities (1:1, 1:N, N:N), and join keys configured between data objects.
• get_semantic_model_logical_view: Retrieves the logical query view for a given model and logical view name.

3. Business Logic & Certified Definitions

• list_semantic_model_measures: Returns certified numerical aggregations (e.g., SUM(Net_ARR) or AVG(Discount)).
• list_semantic_model_dimensions: Returns categorical grouping attributes (e.g., Region, Industry, Fiscal_Quarter).
• list_semantic_model_metrics: Lists enterprise metrics defined with specific business calculation rules.
• get_semantic_model_metric: Inspects the exact formula, time-grain, and filter constraints for a specific metric by modelIdOrApiName and metricIdOrApiName.
• list_semantic_model_calculated_dimensions: Discovers dynamic dimension buckets and conditional classifications.
• list_semantic_model_calculated_measures: Discovers complex calculated measures (such as year-over-year growth percentage).

4. Workspace & General Asset Discovery

• list_workspaces: Lists available analytics workspaces (workspaceIdOrApiName optional).
• list_workspace_assets: Enumerates models, dashboards, and lenses contained inside a specific workspace.
• search_assets: Free-text search across all analytics assets, accepting query and an optional type filter ("dashboard", "visualization", or "semanticModel").

Operational Tips & Troubleshooting#

1. The Initial Client Setup: When connecting an external CLI or MCP client to Tableau Next, ensure Node.js (v18+) is installed if using standard stdio bridges.
2. Session Inactivity Reconnect: External Claude Desktop sessions may drop if idle for more than 15 minutes. Re-initiating the prompt automatically renews the OAuth PKCE handshake.
3. Claude Desktop vs Agentforce Latency: Running Tableau Next queries from Claude Desktop over external HTTPS round-trips averages ~45–60 seconds per turn due to multi-layer API proxying. In contrast, native Agentforce Vibes agents execute queries collocated within Salesforce Hyperforce 10 to 20 times faster.
4. “Unavailable for you at the moment”: If an agent receives this error on analyze_data, check two things:
   • Ensure the user has the Concierge Analytics User permission set assigned.
   • Verify in Tableau Next Setup that the target Semantic Data Model is set to Published and assigned to the user’s workspace.

CRM Analytics MCP Server (analytics/crma-beta)#

For orgs running CRM Analytics (formerly Tableau CRM / Einstein Analytics), the analytics/crma-beta Hosted MCP Server enables AI agents to query pre-aggregated Wave datasets directly using SAQL (Salesforce Analytics Query Language).

Server URLs#

• Standard Global Gateway: https://api.salesforce.com/platform/mcp/v1.0/analytics/crma-beta
• My Domain Route: https://<my-domain>/services/mcp/v1/d/analytics/crma-beta

The 4-Step XMD-Grounded SAQL Workflow#

Never let an AI client guess SAQL field names or run blind queries. CRM Analytics datasets store human-friendly labels, measure formats, and system dimension names inside their Extended Metadata (XMD).

Always instruct your AI client to follow this four-step sequence:

1. Step 1: Discover the Dataset Call list_datasets to locate the target dataset ID (0Fb...).
2. Step 2: Extract the Current Version Call get_dataset with the dataset ID to retrieve currentVersionId (0Fc...). Because Wave dataset schemas can evolve across recipe runs, queries must target the active version.
3. Step 3: Ground on Field Names via XMD Call get_xmd passing the 18-character datasetId, versionId, and xmdType ("main", "system", or "user"). This returns certified field API names, dimension labels, and format masks.
4. Step 4: Execute Schema-Grounded SAQL Call execute_query with a valid SAQL query string referencing the exact dataset and version ID, always terminating with an explicit limit statement.

-- Loaded via execute_query with explicit dataset and version IDs
q = load "0FbB00000001234KAA/0FcB00000005678MAA";
q = group q by ('Account_Owner.Name', 'Region');
q = foreach q generate 
    'Account_Owner.Name' as 'Owner',
    'Region' as 'Region',
    sum('Amount') as 'Total_Pipeline',
    count() as 'Deal_Count';
q = order q by 'Total_Pipeline' desc;
q = limit q 25;

Complete Reference: All 10 CRM Analytics Tools#

Architectural Comparison: SObject vs Data 360 vs Tableau Next vs CRM Analytics#

Frequently asked questions

What is the difference between data/data-cloud-queries and the newer Data 360 MCP Server?

The data/data-cloud-queries server is a dedicated lightweight query server exposing metadata discovery and dynamic SQL execution via post_dc_query_sql. The newer Data 360 server exposes full Connect API capabilities, including data streams, segments, activations, and calculated insights.

How does the Tableau Next MCP Server protect against AI hallucinations in financial metrics?

Tableau Next routes questions through a governed Semantic Data Model rather than raw tables. AI clients call list_semantic_models and get_semantic_model_metric to ground their queries on certified business calculations, dimensions, and measures before calling analyze_data.

Why is British spelling required for the visualization parameter in get_visualization?

The Tableau Next get_visualization tool schema explicitly defines the parameter as visualisationIdOrApiName with an s. Passing visualizationIdOrApiName with a z results in an unrecognized parameter schema validation error.

Why do Claude Desktop calls to Tableau Next feel slower than Agentforce calls?

Claude Desktop communicates with Tableau Next over external HTTPS round-trips via external OAuth gateways, typically averaging around one minute per complex analytical turn. Agentforce runs collocated within the Salesforce Hyperforce mesh, executing analytical queries 10 to 20 times faster.

How do you resolve the Tableau Next error message stating feature is unavailable for you at the moment?

This error occurs when the user or connected app lacks the Tableau Next or Concierge Analytics user permission set, or when the Semantic Data Model has not been published and assigned to the user workspace in Tableau Next Setup.

What is the recommended 4-step workflow for querying CRM Analytics via MCP?

The recommended sequence is: call list_datasets to find the dataset, call get_dataset to extract the 18-character currentVersionId, call get_xmd to inspect field labels and system dimensions, and finally call execute_query with a fully qualified SAQL load statement ending with a limit clause.

Why must CRM Analytics SAQL queries explicitly include the dataset version ID?

In CRM Analytics, dataset schemas evolve across dataflow runs. Passing the exact currentVersionId in the SAQL load statement guarantees that the SAQL compiler resolves the correct field names and metadata snapshots without failing on deprecated or renamed columns.

What is the maximum query timeout and batch size for post_dc_query_sql in Data 360?

The post_dc_query_sql tool enforces standard Data 360 ANSI SQL query execution limits. It returns the initial result batch along with a nextBatchId locator for cursored pagination when queries return large multi-gigabyte analytical result sets.