On May 26, Salesforce announced that the Data 360 MCP Server is now in Developer Preview. The idea is direct: every piece of data in your Salesforce Data Cloud is now reachable by any AI agent that speaks Model Context Protocol. Your CRM context, unified customer profiles, real-time data streams — accessible from Claude Code, Cursor, or any MCP-compatible agent tool without writing a custom API wrapper.
What the Data 360 MCP Server actually exposes
The server gives AI agents structured access to Data Cloud data — unified customer profiles, audience segments, data streams, calculated insights, and identity resolution outputs. These are the data objects that previously required SOQL-like queries against the Data Cloud query engine or custom API development to surface.
Through the MCP interface, an agent can retrieve a unified customer profile by identity, query segment membership for a specific individual, pull calculated insight values for an account, or read real-time data stream events — using the same tool-call pattern it would use to interact with any other MCP server, without learning a proprietary API.
The significance is in the combination. An agent building a renewal recommendation previously had access to Salesforce CRM data — the account record, the opportunity history, the activity log. What it did not have was the Data Cloud layer: the calculated health score from product usage, the segment membership that reflects behavioural patterns, the cross-channel identity resolution that unifies how the same customer appears across touchpoints. The Data 360 MCP Server adds that layer.
Why this changes how agents reason
The practical difference between an agent with CRM access and an agent with CRM plus Data Cloud access is the difference between structured records and contextualised customer intelligence.
An agent reviewing a renewal opportunity can currently see: account name, ACV, contract end date, last activity log, open support tickets. With the Data 360 MCP Server, that same agent can also see: the customer's health score calculated from product usage patterns, their segment membership indicating they are in a high-churn-risk cohort, and their identity resolution confirming that two separate records in the CRM are the same individual.
Because the data arrives through the MCP interface with the same trust layer protections as other Agentforce data access, the agent's data handling governance applies uniformly — no separate security configuration needed for the Data Cloud layer.
What is available in Developer Preview vs. what is coming
| Capability | Available in Developer Preview | Expected at GA |
|---|---|---|
| Unified customer profile retrieval | ✓ Query by identity, return full profile with attributes | — |
| Audience segment membership | ✓ Query segment membership for a specific individual or account | — |
| Calculated insights reads | ✓ Return health scores, propensity scores per record | — |
| Identity resolution queries | ✓ Cross-reference unified identity across touchpoints | — |
| Data stream reads | ✓ Basic event stream data — subject to Data Cloud sync latency | Real-time streaming subscriptions |
| Write-back to Data Cloud | ✗ Not yet available | Agents will be able to update Data Cloud records based on reasoning output |
| Complex data stream subscriptions | ✗ Not yet available | Subscribe to data stream events as part of agent trigger logic |
| Production org access | ✗ Developer Edition orgs with Data Cloud only | Full production org access at GA |
Developer Preview gives access to the core unified profile retrieval, segment queries, and calculated insight reads. The key constraint is data freshness: Data Cloud data surfaces through the MCP Server with the same refresh latency as the underlying Data Cloud sync. For most use cases this is acceptable. For agents reasoning about real-time events, it is worth understanding the lag characteristics of your specific data streams before building workflows that depend on sub-second freshness.
Developer Preview access path
Developer Preview is the right time to experiment, not the right time to build production workflows. Map your architecture, test your data access patterns, and identify what works before GA removes the preview caveats. The orgs that move through this now will deploy faster when GA lands.
