While Salesforce native capabilities like Agentforce Vibes and Agentforce Coworker are steadily improving and adding capabilities, many customers still prefer to bring their own AI harnesses like Claude Code, ChatGPT, Cursor, etc. and connect them to Salesforce using MCP for queries and configuration. This is mainly due to the excellent UX and performance provided by these tools, which is difficult to reproduce through employee agents or Salesforce Vibes.

Also, while AI agents could use the sf CLI or Connect API those are not designed to be effectively leveraged by an AI agent, often blowing up its context window or requiring trial and error while learning how to use the CLI or API. The MCP implementation is designed with progressive disclosure (meaning only revealing details on-demand, as needed) and optimal outputs that should result in better, more efficient interactions.

• Data 360 MCP Server. NOTE: for a comparison of Salesforce-hosted MCP servers and their functionality, see MCP Solutions

In addition to these MCP servers, consider using these agent skills to teach your agents the optimal coding and configuration workflows curated by Salesforce.

IMPORTANT NOTE ABOUT TRUST AND PRIVACY

As mentioned above, many customers may prefer using their favorite AI services for interacting with the platform headlessly rather than with Salesforce-native capabilities like Vibes and Coworker, but the Agentforce-branded Salesforce capabilities still have the Salesforce Trust Layer. If you are going to bring your own client, ensure that you have an enterprise agreement that includes similar capabilities for protecting your business data, such as Claude for Enterprise ChatGPT for Enterprise, Amazon Bedrock, etc. and do not utilize these MCP capabilities with unauthorized personal AI subscriptions which usually have looser agreements around data privacy and retention.