Product1 publisher2 min readPublished
Salesforce pipes live CRM actions into Amazon Quick and Gemini Enterprise over MCP
At Dreamforce 2026 the company said Model Context Protocol will carry its context and actions into both hyperscalers' agent products, while AWS agents move into Slack. The connector work becomes a scoping job.
The Product Desk · Product desk
What happened
- Salesforce used Dreamforce 2026 in San Francisco to announce expanded integrations with both AWS and Google Cloud, moving CRM data, agents and workflows into its partners' cloud environments.
- Salesforce business context and actions are available natively inside AWS' Amazon Quick agent through the Model Context Protocol, so Quick can read live CRM data and act for sales and customer service teams.
- AWS frontier agents start arriving in Slack with the AWS DevOps Agent, which Salesforce said lets teams resolve infrastructure problems without leaving the messaging tool.
- Through a new headless architecture built on MCP, Google's Gemini can run sales pipeline health checks, account summaries and case triage across sales, service, finance and advertising workflows.
Compiled by The Product DeskSomething wrong?How this is made
Why it matters
- decision Platform teams now choose which product a given role should start an agent job in, because three surfaces reach the same CRM record and the connector no longer decides it for them.
- exposure A write path into CRM records sits inside a product Salesforce does not administer, so whoever answers for a bad record change has to trace it through an AWS console too.
- constraint Customers who must certify the infrastructure under their Salesforce org get that review on a fourth-quarter timetable set by the vendor's migration plan.
- precedent Both hyperscalers got their access through MCP, which puts the protocol itself, not a Salesforce-built connector, in front of the security reviewer for every agent vendor that comes next.
A sales ops lead who finds an opportunity edited on Friday afternoon by nobody in particular now has more than one console to check. Autonomous work on CRM records can originate inside an AWS product [2], and the model behind an Agentforce agent can now come from Amazon Bedrock's library, including Anthropic, OpenAI and Nvidia [4].
Most of the AWS capability on this list lands inside products Salesforce controls. Amazon Quick reaching Salesforce context is one item of five [2]. The other four run the other way: AWS frontier agents in Slack beginning with the DevOps Agent [3], the Bedrock model library inside Agentforce [4], Data 360 Zero Copy extended to AWS Glue-managed Apache Iceberg tables and Amazon Aurora [5], and bidirectional audio for Agentforce users through Amazon Connect over Websockets [6][1].
What MCP removes is the connector build. Salesforce's headless architecture, also built on MCP, is what lets Gemini reach CRM capability without the customer doing integration work [10][11]. The scope decision stays put. Someone has to say which actions an outside agent may take, on which objects, for which roles, and SiliconANGLE's account does not say how those permissions are scoped.
AWS Vice President of Data and AI GTM Rahul Pathak said the integrations "give customers data access without migration, model choice at the right economics and AI agents where their people already work" [7]. For a lot of enterprises the place people already work is Slack, and the same announcement sends an AWS agent into it [3].
The Google Cloud half of the announcement moves infrastructure. Salesforce workloads run on Google Cloud through Hyperforce, and the company said Hyperforce there is already supporting live customer traffic, with broader customer migrations expected in the fourth quarter [8]. "What changes here is the infrastructure underneath our customers, not the trust they've built with us," said Salesforce President and Chief Trust and Infrastructure Officer Meir Amiel [9].
The sorting question for each agent job on a roadmap is which product the person starts in and which system owns the record that changes. Where those sit with the same vendor, the supplied connector is enough and the permission model already in place covers it. Where they differ, say a rep starting in Amazon Quick and writing to a Salesforce opportunity, the job needs a named scope and one owner who can still read the log in six months. Three agent surfaces can now reach the same CRM record [2].
What to watch
- Whether Salesforce publishes the permission scopes and audit trail for MCP-exposed CRM actions taken in Amazon Quick and Gemini Enterprise.
- Whether the broader Hyperforce migrations to Google Cloud land in the fourth quarter as Salesforce expects.
- Which AWS frontier agents follow the DevOps Agent into Slack, and whether any of them write to Salesforce records.