Integrations · CRM
Salesforce AI integration for Sales Cloud, Service Cloud and Slack
Agentforce is Salesforce's own AI. We connect Claude to the same org through Salesforce's hosted MCP servers or its APIs, for jobs your fields and rules define. Which writes a person signs off is up to you.
How does a Salesforce AI integration work?
A Salesforce AI integration from AILoopwise connects Claude to your org through Salesforce's hosted MCP servers or its APIs, once your admin creates an External Client App and enables each server. Claude works under each user's permissions in Sales Cloud and Service Cloud, drafts updates and replies, and you set which writes a named person approves.
Where things stand
- As of 2026-11-04, Salesforce's hosted MCP servers are generally available, name Claude among supported assistants, and stay off until an admin enables each.1,2
- As of 2026-11-04, Salesforce in Claude is a beta plugin for orgs Salesforce approves on the latest Sales Cloud enterprise edition; it asks before each write by default.6
- Salesforce's press release of 2026-08-27 says Claude is available in Agentforce as a reasoning model for the Atlas Reasoning Engine, through Amazon Bedrock.7
- As of 2026-11-04, Salesforce calls Data 360 the official new name for what was previously Data Cloud.18
What we work on in Salesforce
Sales Cloud
Salesforce's sales application: Leads, Accounts, Opportunities and forecasts; its built-in AI features include lead scoring.
Keeping Stage, Close Date and Next Step true to the activity history, drafting follow-ups, and filling the custom qualification fields your admin added.
Service Cloud
Salesforce's customer service application: Cases, queues, Knowledge, and assignment, escalation and auto-response rules.
Summarising a Case before hand-off, drafting a reply from Knowledge for the agent to edit, and spotting misrouted Cases.
Marketing Cloud
Salesforce's marketing suite: Marketing Cloud Next, Marketing Cloud Engagement, and Marketing Cloud Account Engagement (formerly Pardot) for B2B.
Turning a brief into segment criteria, checking draft emails against brand rules, and preparing Campaign Members for sales.
Agentforce
Salesforce's product for building and running AI agents: the Atlas Reasoning Engine plans each request; Agent Builder assembles agents from Flows, prompts, Apex and MuleSoft APIs.
Salesforce's packaged jobs: prospecting, outreach, Pipeline Management, briefs, and hand-off to a person when a request is out of scope.
Data 360
Salesforce's data layer, formerly Data Cloud: one customer profile unified from many sources, read in place through Zero-Copy.
Giving an agent the whole customer profile, not one object, as context for a summary.
Slack
Salesforce's messaging platform and, per Salesforce, the conversational interface for Agentforce. Slack AI summarises channels and threads.
Posting a deal or Case digest where the team already talks, and answering a question about an Opportunity in a thread.
Tableau
Salesforce's analytics platform, including Tableau Next.
Answering a pipeline or service question from a published data source and drafting the weekly narrative.
Where AI helps in Salesforce
Lead capture and routing
LeadWeb-to-LeadQueueAccount
- When
- A Lead arrives from a Web-to-Lead form or lands in an unassigned queue
- Reads
- the form fields, Lead Status, any existing Account and owner, and your routing criteria
- AI
- checks fit against your criteria, finds the Account owner, and drafts a first reply with the routing reason
- Result
- a proposed owner or queue and a draft reply on the Lead; nothing is sent
- Approved by
- the SDR or sales rep who owns the queue24,29,4
Record quality and duplicates
LeadContactAccountmatching rulesduplicate rules
- When
- Weekly, or when a duplicate rule alerts instead of blocking
- Reads
- records your matching rules flagged, their owners and related records, and empty required fields
- AI
- groups likely duplicates, proposes the surviving record and values, and lists empty fields by owner
- Result
- a merge proposal per group and a fix list per owner
- Approved by
- the Salesforce admin for merges; record owners for field fixes23
Follow-up and email drafts
OpportunityContactEventEinstein Activity Capture
- When
- A meeting Event on an Opportunity ends, or Einstein Activity Capture logs the buyer's reply
- Reads
- the activity timeline, the Opportunity's Stage and Next Step, and the Contacts on the deal
- AI
- drafts the follow-up email and a new Next Step, quoting the note or email behind each
- Result
- a draft in the rep's Outlook or Gmail and a proposed field change; no email leaves
- Approved by
- the Opportunity owner26,12
Pipeline and forecast review
OpportunityTaskEventFlow
- When
- A scheduled Flow the day before the forecast meeting
- Reads
- open Opportunities closing this period: Stage, Close Date, Next Step, Forecast Category, last activity
- AI
- flags records whose Stage or Close Date contradicts the activity, with evidence
- Result
- one review list per rep, one proposed change per line
- Approved by
- the Opportunity owner; the sales manager sees what was declined12,27
Case summaries and reply drafts
CaseEmailMessageQueueKnowledge
- When
- A Case is escalated or moved to another queue
- Reads
- the email thread, Case history, Account and matching Knowledge articles
- AI
- writes a short summary for the next agent and drafts a reply naming the Knowledge article used
- Result
- a summary on the Case and a reply draft for the agent
- Approved by
- the service agent who owns the Case25,14,4
How we connect
Four routes to AI automation in Salesforce. Your edition, licences and admin decide which are open to you.
| Route | What it is | Fits when | Limits |
|---|---|---|---|
| Built inAgentforce | Salesforce's own agents for prospecting, outreach, Pipeline Management on Stage and Next Steps, briefs, quoting and service replies. | you license it and its standard jobs match how your teams sell and serve | AI can be added to Core and above; Pipeline Management needs Einstein Activity Capture and Conversation Intelligence.8,9,10,15 |
| Vendor MCPSalesforce Hosted MCP Servers | Salesforce-run servers for Claude web and Claude Desktop: platform/sobject-reads (read-only), platform/sobject-all (read, write, delete) and custom tools from Apex or Flows. | Enterprise Edition or above, with Claude on your records under each user's permissions | admin setup and the daily API quota; see the limits below.1,2,3,4,5 |
| Vendor MCPSalesforce in Claude (plugin, beta) | 37 sales skills built by Salesforce, bundled with the Salesforce Claude connector and Slack's connector in Claude chat and Cowork. | your reps want to ask and update Salesforce from Claude, and Salesforce has accepted your org into the beta | beta and gated; it acts when a user asks, not on a record event.6,7 |
| API buildA build on your org | Our service, started by a Flow, reading through the read-only server or Named Queries and writing only through custom tools that enforce your rules. | the job runs on a record event, crosses Sales Cloud, Service Cloud and Slack, or follows a policy only your org has | your admin sets up the External Client App and tools with us; each approval step has a named owner.4,27 |
Claude already runs inside Agentforce as a reasoning model. If you license Agentforce and its Pipeline Management and service replies cover your fields and rules, Agentforce alone is enough. A Claude build earns its place where Agentforce is not licensed, where the job runs on your custom objects, or where it crosses Sales Cloud, Service Cloud and Slack on a record event. A rep who only asks and updates needs just the plugin, once its beta covers your org.
AILoopwise is Certified in the Claude Partner Network.
Is this a fit?
A good fit
- Teams on Sales Cloud Enterprise Edition or above whose reps skip field updates
- Service Cloud teams that rewrite the same Knowledge answer every day
- Orgs whose custom objects and approval rules Agentforce is not set up for
- Admins who want AI under existing permissions, starting read-only
Not a fit
- Agentforce is licensed and already fills the fields and replies you work by
- Starter Suite or Pro Suite with no API access on the contract: settle the edition first
Limits and data handling
- Hosted MCP servers require Enterprise Edition or above (Developer Edition for testing) and stay disabled until an admin enables each one.2
- Authentication runs only through an External Client App with the mcp_api and refresh_token scopes; Connected Apps are not supported, and a new app can take up to 30 minutes to work.3
- MCP tool calls consume the org's daily API quota, so we start jobs from record events, not polling.2
- Claude sees only what the signed-in user may see: object permissions, field-level security and sharing all apply.4
We agree with you, per project, where each part of the system runs and who can access the data, typically inside your own systems and accounts. When we use Claude through Anthropic's API, the model call runs at Anthropic and is governed by contract.
How a project runs
- 01
Find the loop We sit with your team and list the work that comes back every week. Then we pick the loop that costs the most hours.
- 02
Map it Every step, every tool, every decision: where the inquiry lands, who touches it, what they look up, where the time goes.
- 03
Build the agents Each step gets an agent with the skills to do it inside your tools. One reads and sorts every inquiry, one logs it in the CRM, one pulls up the customer's history, one drafts the reply and the follow-ups.
- 04
Add the checkers A second set of agents checks every result: wrong data, a duplicate, a missed follow-up. Anything unclear goes to a person.
- 05
Close the loop The loop runs. Your team approves, handles the exceptions and talks to the customer.
Step 2 includes your Salesforce admin, who creates the External Client App and enables each server.
Questions about Salesforce
- How do I use AI in Salesforce?
- You use AI in Salesforce in one of three ways: Agentforce, Salesforce's own AI agents; the Salesforce in Claude plugin, for reps who work from Claude; or a Claude build on the hosted MCP servers. Agentforce fits its standard jobs if you license it. Custom objects, your own rules and jobs across clouds need a build with an approval step.
- What AI features does Salesforce have?
- Salesforce's AI features sit in Agentforce and in each cloud. Agentforce in Sales Cloud builds prospect lists, drafts outreach, updates Opportunity Stage and Next Steps, and writes briefs. Sales Cloud adds AI lead scoring, Service Cloud adds Conversation Summaries and AI-generated replies, and Einstein Activity Capture logs emails and events.
- Does Salesforce have an MCP server?
- Yes. Salesforce runs hosted MCP servers, generally available as of 2026-11-04, and its own repository names Claude and ChatGPT among the supported assistants. platform/sobject-reads is read-only; platform/sobject-all can also create, update and delete. They need Enterprise Edition or above, and an admin must enable each server before any client connects.
- How do I connect Salesforce to Claude?
- A Salesforce Claude integration runs through the hosted MCP servers, the Salesforce in Claude plugin, which bundles the Salesforce Claude connector, or a build on your org. Each starts with your admin creating an External Client App; for hosted servers the admin also enables each server in API Catalog. Claude then works with each user's own permissions.
- Salesforce MCP vs Agentforce: which do I need?
- Often both. Agentforce runs Salesforce's own agents inside the org, and Salesforce says Claude is available there as a reasoning model through Amazon Bedrock. Hosted MCP lets Claude read records and call custom tools your admin builds from Flows or Apex. Choose MCP where Agentforce is not licensed or not set up for your fields.
- Can Claude change Salesforce records without asking?
- In our builds, not by default: Claude proposes and the record owner approves, unless you decide a step can run on its own. The Salesforce in Claude plugin asks before each write by default, per Anthropic's help article. On hosted servers we start read-only and route writes through custom tools your admin approves, so Salesforce itself enforces a rule such as the allowed Case statuses.
- How much does Salesforce Agentforce cost?
- Salesforce sets that in your contract; we do not quote its prices. Its Agentforce pricing page lists the models: Flex Credits charged per action, per-conversation pricing for customer-facing agents, per-user licences and add-ons. We price a Claude build separately; its cost drivers are the number of jobs, the custom tools needed and the approval steps.
- Does this work on Salesforce Starter Suite or Pro Suite?
- Partly; the edition decides. Hosted MCP servers need Enterprise Edition or above, and Salesforce's pricing page does not map that to a suite, so we check your edition with your admin first. The page lists Web Services API as an additional line. Without API access, Claude works on an export and people update the records.
Related integrations
Sources
- github.com/forcedotcom/mcp-hosted (read 2026-11-04)
- github.com/forcedotcom/mcp-hosted/wiki/Confirming-Availability (read 2026-11-04)
- github.com/forcedotcom/mcp-hosted/wiki/Configuring-an-External-Client-App (read 2026-11-04)
- github.com/forcedotcom/mcp-hosted/wiki/Security-and-Permissions (read 2026-11-04)
- github.com/forcedotcom/mcp-hosted/wiki/FAQ (read 2026-11-04)
- support.claude.com/en/articles/16952184-set-up-salesforce-in-claude-for-your-organization (read 2026-11-04)
- www.salesforce.com/ap/news/press-releases/2026/08/27/sg-salesforce-and-anthropic-announce-claudeforce-the-1-ai-meets-the-1-ai-crm/ (read 2026-11-04)
- www.salesforce.com/sales/ai-sales-agent/ (read 2026-11-04)
- trailhead.salesforce.com/content/learn/modules/build-an-agent-to-manage-your-pipeline/enable-agentforce-pipeline-management (read 2026-11-04)
- www.salesforce.com/sales/pricing/ (read 2026-11-04)
- www.salesforce.com/sales/ (read 2026-11-04)
- trailhead.salesforce.com/en/content/learn/projects/customize-a-sales-path-for-your-team/customize-opportunity-stages (read 2026-11-04)
- www.salesforce.com/company/legal/tmcusageguidelines/ (read 2026-11-04)
- www.salesforce.com/service/ (read 2026-11-04)
- www.salesforce.com/service/ai/ (read 2026-11-04)
- www.salesforce.com/marketing/ (read 2026-11-04)
- www.salesforce.com/agentforce/ (read 2026-11-04)
- www.salesforce.com/data/ (read 2026-11-04)
- slack.com/ai (read 2026-11-04)
- docs.slack.dev/ai/mcp-server/ (read 2026-11-04)
- github.com/tableau/tableau-mcp (read 2026-11-04)
- www.salesforce.com/analytics/ (read 2026-11-04)
- trailhead.salesforce.com/content/learn/modules/sales_admin_duplicate_management/sales_admin_duplicate_management_unit_1 (read 2026-11-04)
- trailhead.salesforce.com/content/learn/modules/prospect-tracking-with-salesforce-leads-quick-look/track-prospects-with-salesforce-leads (read 2026-11-04)
- trailhead.salesforce.com/content/learn/modules/service_lex/service_lex_case_manage (read 2026-11-04)
- trailhead.salesforce.com/content/learn/modules/email-and-calendar-integrations-quick-look/learn-about-email-and-calendar-integration-tools (read 2026-11-04)
- trailhead.salesforce.com/content/learn/modules/record-triggered-flows/add-a-scheduled-task-to-your-flow (read 2026-11-04)
- trailhead.salesforce.com/content/learn/modules/campaign_basics/campaigns_basics_unit_3 (read 2026-11-04)
- trailhead.salesforce.com/content/learn/modules/lead-qualification-with-scoring-and-grading/support-a-lead-qualification-model-with-account-engagement (read 2026-11-04)
- trailhead.salesforce.com/content/learn/modules/outlook_integration/outlook_integration_unit_3 (read 2026-11-04)
- trailhead.salesforce.com/de/content/learn/modules/sales_admin_duplicate_management/sales_admin_duplicate_management_unit_1 (read 2026-11-04)
- trailhead.salesforce.com/de/content/learn/modules/service_lex/service_lex_case_manage (read 2026-11-04)
- www.salesforce.com/agentforce/pricing/ (read 2026-11-04)
- www.salesforce.com/artificial-intelligence/ (read 2026-11-04)
No company should leave its growth untapped.
See how much potential is in your organization.
We only take on companies where we see clear growth potential. In a 30-minute assessment we find out whether that applies to you. If it doesn't, we'll tell you.