Solution · Claude + Salesforce

Claude + Salesforce: connect, automate, and ship AI workflows in your org

Salesforce's Hosted MCP servers made the connection possible. Making it safe, governed, and genuinely useful is the consulting problem — and we solve it from an unusual seat: GAT built and operates ApexGenius, a production Salesforce MCP for working with live org data and metadata from Claude or ChatGPT.

  • 50+ Salesforce projects
  • 100+ integrations
  • 5.0 client rating
  • 6+ years Salesforce
Abstract illustration: two hexagonal network clusters joined by luminous azure threads

Plain English first

What MCP actually is

The Model Context Protocol (MCP) is the open standard that lets an AI assistant like Claude connect to real systems — read data, call actions, respect permissions — instead of guessing from training data. Think of it as the USB port between Claude and Salesforce.

In April 2026, Salesforce shipped Hosted MCP servers and a first-party path for Claude. That changed the question from “can we connect AI to our org?” to “which of the four ways should we, and what do we let it touch?”

That second question is where teams get stuck — and where we live.

The how-to

Hosted MCP setup, step by step

The first-party route. Roughly an afternoon for an admin who knows their org — faster with us on the call.

  1. 01

    Confirm your edition & permissions

    Salesforce Hosted MCP servers are available on supported editions. You'll need admin access to create connected/external client apps and assign permission sets.

  2. 02

    Create the External Client App

    In Setup, create the app that will broker Claude's OAuth connection — this is where scopes and policies get pinned down, not after the fact.

  3. 03

    Activate the hosted MCP servers

    Enable the MCP servers you actually need. Start narrow: fewer surfaces, fewer surprises, faster security review.

  4. 04

    Add the connector in Claude

    In Claude's settings, add Salesforce as a connector and complete the OAuth flow as a user whose profile reflects the access you intend AI to have.

  5. 05

    Test in a sandbox, then govern

    Run your first workflows against a sandbox. Confirm FLS and sharing behave as expected, then write down the rules of engagement before production.

Prefer the two-minute version? ApexGenius connects Claude (and ChatGPT) to any org without the app setup — it's the product we built for exactly this.

The honest comparison

Four ways to connect AI to Salesforce

Product list pricing shown for orientation — verify current terms with each vendor. What we sell is the judgment, not a subscription.

Comparison of Salesforce AI connection options
 Salesforce Hosted MCPAgentforce / BYOLLMApexGeniusCustom MCP build
Setup timeAn afternoon (admin required)Weeks — platform project~2 minutes per orgWeeks of engineering
Works withClaude (first-party path)Inside Salesforce + ChatGPT surfacesClaude & ChatGPTAnything you build for
Read / write scopeData + selected actionsPlatform-native agents & actionsFull read/write, incl. metadataWhatever you implement
Deploy metadata / codeLimitedNo — different jobYes — Apex, LWC, FlowIf you build it
Governance modelYour org's OAuth + FLS/sharingSalesforce-managed, deepest native controlsOAuth user context + org permissionsYou own every decision
Cost model (list)Included with supported editionsPer-user add-on + usage (Flex Credits)Flat product subscriptionEngineering time

Governance

Your security model still applies — if you set it up right

MCP doesn't bypass Salesforce security. Claude sees what the connected user can see. That's reassuring — and exactly why the connected user is a design decision, not an afterthought.

Least-privilege by design

A dedicated permission set for AI access — scoped objects, scoped fields, no blanket Modify All. We design it with your admin, in writing.

FLS & sharing, verified

Field-level security and sharing rules govern every MCP read. We test the boundaries in a sandbox before production sees a single prompt.

PHI & regulated data

For healthcare and financial teams: minimum-necessary access, audit trails, and vendor-agreement review with your compliance team before rollout.

Monitoring & revocation

Connected-app usage is visible and revocable. We instrument who's connected, what's queried, and how to shut it off in one click.

Healthcare team? See how we handle Salesforce + AI in clinical settings →

What to ship first

Workflows by role, not by demo

Admins

  • Describe → build: flows, fields, and validation drafted conversationally
  • Org Q&A without clicking through Setup
  • Debug faster with AI that can actually see the metadata

RevOps

  • Pipeline hygiene sweeps before the forecast call
  • SOQL-free answers to revenue questions
  • Data quality triage with proposed fixes, not just alerts

Support & service

  • Case summarization with next-step drafts
  • Knowledge lookups grounded in your org's data
  • Escalation digests that write themselves

Developers

  • Apex written, reviewed, and tested in conversation
  • Deploys and metadata changes through MCP (with ApexGenius)
  • Release analysis and test planning before metadata changes ship

GAT Solutions helped build the Salesforce system almost entirely from scratch, with thoughtful solution design, clear communication, and dependable delivery.

Greenfield Salesforce system · Verified client review

Questions

Frequently asked questions

Is it safe to connect Claude to Salesforce?
Yes, when governed properly. MCP connections run through OAuth in a user context — field-level security, sharing rules, and permission sets all still apply. The risk isn't the protocol; it's connecting a over-permissioned user. We design least-privilege access before anything touches production.
What Salesforce editions support Hosted MCP servers?
Salesforce ships Hosted MCP on supported modern editions; availability and included capacity vary by edition and evolve quickly. We confirm your org's exact entitlement during the consultation — and if Hosted MCP isn't available to you, ApexGenius works with any org.
How is this different from Agentforce?
Agentforce builds autonomous agents that live inside Salesforce — a platform project with platform pricing. Claude over MCP gives your team an AI assistant that works with Salesforce from the outside. Many teams need the assistant long before they need the platform. We advise on both honestly — see our Agentforce consulting service.
Can Claude actually change things in the org — or just read?
Hosted MCP supports data operations and selected actions. For metadata — writing Apex, deploying Flows, creating fields — that's what we built ApexGenius for: full read/write including deploys, in both Claude and ChatGPT.
What does the setup cost?
The connection may be covered by an existing Salesforce entitlement or use a separate flat product subscription such as ApexGenius. AI-client, Salesforce usage, and consulting costs remain separate. See the ApexGenius product site for current pricing, or book a consultation for an implementation scope.
What should we ship in the first month?
One workflow per team, chosen for visible time savings: a pipeline-hygiene sweep for RevOps, case summarization for support, and describe-to-build for admins are the usual first three. Small, governed, measured — then expand.

Initial systems consultation

Connect Claude to Salesforce — governed, useful, this month.

  • You leave with
  • A focused consultation with a senior systems consultant
  • A current-state fit and architecture read
  • A recommended path — implementation, ongoing ownership, or product