Part Two: What a Claude Salesforce Integration Looks Like Across Every App
This is part two of a three-part series. Part one made the case for keeping your systems of record instead of rebuilding them with AI. This part shows what a Claude Salesforce integration looks like when it reaches every app you run. Part three will show it working, live, in a full customer demo.
You took the advice. You did not rip out Salesforce, and nobody vibe coded a replacement over a long weekend. Good.
Now look at how your team actually works. Open Salesforce. Then the ERP. Then the inbox. Then the shared drive. Copy a number from one screen into another until the picture makes sense. Repeat tomorrow.
Here is the next move. Salesforce is the spine. Claude is the cockpit. Your systems keep holding the facts, the rules and the permissions. Claude becomes the one seat where you see across all of them, ask a question in plain English, and approve the next step. You are not adding another app. You are giving every app you already own one place to report to.
The short version
- Spine and cockpit. Salesforce and your other systems of record hold the data, rules and permissions. Claude is where a person sees across them and decides.
- Salesforce is building for this. Its AIforce strategy brings Salesforce to where people already work, including Claude, Slack and ChatGPT.
- Agentforce is the autopilot. It runs governed, high-volume work inside Salesforce. The cockpit and the autopilot make each other better.
- The value is in the seams. Real workflows cross four or five systems. The cockpit reaches all of them without moving the data.
- A person stays in the pilot’s seat. Claude inherits each system’s permissions and drafts the next step. You approve it.
Salesforce Is Already Bringing the CRM to Where You Work
The clearest endorsement of this architecture came from Salesforce itself. At Dreamforce 2026, Salesforce introduced AIforce with a simple thesis: people should not have to come to Salesforce to get work done, so Salesforce comes to them.
- Claudeforce puts Salesforce inside Claude, with 37 prebuilt sales skills that take governed action in your org.
- The Headless Toolkit exposes Salesforce through MCP servers, APIs and skills to any interface you build against, including Teams and ChatGPT.
- Every action still runs inside your existing permissions and business rules.
The plumbing underneath is just as settled. The Model Context Protocol (MCP) has become the standard way AI connects to business systems. Salesforce, ServiceNow, SAP, Microsoft, Snowflake, Databricks, Box, Intuit and Stripe all ship official servers. Salesforce’s hosted MCP servers went generally available in April 2026, and field level security and sharing rules apply to every call.
We run Cirrius this way. Our CEO starts every day from a Claude cockpit connected to Salesforce, our accounting system, email, calendar and meeting notes.
Five Roles in a Claude Salesforce Integration
A metaphor only helps if every part has a clear job. Here is how the pieces fit.
Salesforce: the spine
Holds the customer record, the rules and the permissions.
Hover for detail
Salesforce: the spine
Years of opportunities, cases, quotes and pricing logic, wrapped in field level security and sharing rules. Everything else reads from it and writes back to it.
Your other systems: the instruments
ERP, accounting, service, documents, email and meetings.
Hover for detail
Your other systems: the instruments
Each one is the source of truth for its slice of the customer story. The cockpit reads them in place through governed connectors. Nothing gets copied into a new database.
Agentforce: the autopilot
Runs governed, high-volume work inside Salesforce.
Hover for detail
Agentforce: the autopilot
Service conversations, lead qualification, case routing. Work you have defined well, running at scale with a full audit trail. Prove a workflow in the cockpit, then let Agentforce fly it every day.
Claude: the cockpit
One seat that sees across every system.
Hover for detail
Claude: the cockpit
Ask in plain English and get a sourced answer that pulls from Salesforce, the ERP and your inbox at once. Claude drafts the next step: the email, the update, the quote change.
You: the pilot
A person approves every move that matters.
Hover for detail
You: the pilot
Claude can only reach what you are allowed to reach, and it asks before it writes. You keep the judgment. The cockpit takes away the clicking.
Notice what stays at the center. The cockpit does not replace the spine. It makes the spine worth more, because something finally reads all of it.
“Frontier models depend on CRM. They don’t replace it.”
Marc Benioff, Chair and CEO, Salesforce, August 2026
A Day in the Cockpit
Here is what this looks like for a sales leader at a mid-market company. Nothing below is science fiction. Every step uses connectors that exist today.
- 7:00 AM. The brief is waiting. New inquiries, quotes waiting on a reply, accounts that went quiet, and the three calls that matter most today. Pulled overnight from Salesforce, the inbox and the calendar.
- 9:30 AM. “Give me the full picture on this customer.” One answer: open opportunities from Salesforce, order history from the ERP, the open invoice from accounting, and what was said on last week’s call. Every fact cites its source.
- 11:15 AM. Speed to lead. A web inquiry lands. Claude checks for an existing account, drafts a personal reply and suggests a meeting time. You change one line and approve.
- 2:00 PM. The question that used to be a report request. “Which customers bought less this quarter than last, and do any of them have open service cases?” That crosses three systems. It used to be a ticket to IT. Now it is a sentence.
- 4:45 PM. Close the loop. Claude proposes the Salesforce updates from the day’s calls and emails: next steps, stage changes, follow up tasks. You approve the batch, and the record is current before you log off.
Look at what happened. The CRM got more complete on a day nobody opened it to type. That is the argument from part one in practice: agents make the system of record more valuable, because now something reads and maintains every field.
Why the Cockpit Has to Span Every App
The customer’s story is spread across your stack. The opportunity is in the CRM. The order and invoice are in the ERP. The payment is in accounting. The contract is in a document store. One workflow, like quote to cash, touches four or five of those systems before it produces any value.
That is why a cockpit inside a single application solves the wrong problem. The real cost lives in the seams, in the swivel chair work where a person copies a value from one screen into another. A cockpit that spans every system changes that:
- Data stays where it lives. No migration and no new warehouse to govern. Every major vendor’s data strategy now federates in place instead of moving it.
- Each system’s permissions come along. Claude connects as an authenticated user, so it sees what that person can see and nothing more.
- It grows one connector at a time. Start with Salesforce and email. Add the ERP next quarter. Every new connection makes every question smarter.
- Your options stay open. MCP is an open standard, so the systems you connect today work with the models you choose tomorrow.
What Keeps a Claude Salesforce Integration Safe
Executives ask the right question first: who is accountable when the AI gets it wrong? The architecture answers it.
- The agent cannot exceed the person. Object permissions, field level security and sharing rules apply to every Claude call into Salesforce, exactly as they apply to the user.
- Reads first, writes with approval. Start read-only. When Claude begins drafting updates, a person approves them before they land.
- Integration identity is the new perimeter. The 2025 incidents that hit Salesforce customers came through stolen third party tokens and rogue connected apps, not platform flaws. Least privilege integration users, connected app allowlists and token rotation close that door.
- Your data stays yours. Salesforce committed to zero data retention with the model provider, and enterprise Claude plans do not train on your data.
- The record has to be clean. As Forrester puts it, if data is messy, AI will scale the mess. Fix the record before you connect to it.
Where to Start
Pick one cross-system workflow, not one app. The morning brief, or quote to cash. Something that crosses at least three systems and that someone does by hand today.
Clean the record it depends on. Required fields, duplicates, ownership. The cockpit will read every field, so every fix compounds.
Connect read-only through governed connectors. Least privilege integration users and official MCP servers. Prove the answers are right before anything writes.
Turn on writes with a person approving, then measure. Track hours returned and record completeness. When a workflow is stable and high volume, hand it to Agentforce to run at scale.
Claude Salesforce Integration: Common Questions
Do I need Claudeforce to connect Claude to Salesforce?
Claudeforce is the fastest path for sales teams. It ships prebuilt sales skills that take governed action in your org and activates through AppExchange. Salesforce’s hosted MCP servers also let Claude reach your org with your permissions intact. Most companies start with Claudeforce and add connectors for the rest of their stack.
Does Claude replace Agentforce?
No. They do different jobs. Agentforce runs governed, repeatable work inside Salesforce at volume. Claude is where a person works across every system. The strongest setups use both: prove a workflow in the cockpit, then let Agentforce run it.
Can Claude connect to systems other than Salesforce?
Yes. That is the whole point of the cockpit. ERP, accounting, service desks, document stores, email and calendars all connect through MCP servers or APIs, each with its own permissions. Claude answers across all of them in one conversation.
Does this only work with Claude?
No. We lead with Claude because Claudeforce puts Salesforce inside it natively, and because we have run our own business through it for more than a year. But the architecture is the point, not the brand. We build this pattern across Claude, OpenAI, Agentforce and Salesforce, and Salesforce’s new Agent Fabric governs agents from multiple vendors in one registry. Pick the cockpit your people will actually sit in.
How long does it take to get started?
Most engagements reach a Production Readiness checkpoint at Day 30 and a live agent your team owns by Day 90. Priorities and capacity are set with you up front, so the pilot cohort stays deliberately small.
We Fly This Every Day
Cirrius has implemented Salesforce since 2009, runs 50+ Claude agents in production, and was named 2027 Salesforce Partner of the Year for Customer Value. We connect the spine, build the Claudeforce cockpit, and hand your team the controls.
Part three is next: a working demo of the full cockpit, plus the lessons we learned building it.
Want to see what your cockpit could look like? Talk to our team.

