Claude for Salesforce in 2026: every route, what it costs, and who it fits

Rama Adi
Written by

Rama Adi

Katelin Teen
Reviewed by

Katelin Teen

Last edited November 18, 2026

Expert Verified
Hand-drawn illustration of a seller, a manager and a support agent working with Claude on Salesforce account and case records
TL;DR

Claude for Salesforce now mostly means Claudeforce, the partnership Salesforce and Anthropic announced on August 26, 2026. The headline product, Salesforce in Claude, is a beta plugin with 37 skills, all of them for sellers, and it only opens to orgs Salesforce approves. Sales teams should apply. For support, the answer is narrower: Claude can read and update a Case through Salesforce's hosted MCP servers, but nothing Anthropic or Salesforce ships yet answers your customers, and those MCP calls are about to carry their own Salesforce meter.

For case work, eesel works inside your Salesforce org as a support teammate, and you can talk to it from Claude.

What does "Claude for Salesforce" mean after Claudeforce?

It means a lot more than it did in the summer. On August 26, 2026, Salesforce announced Claudeforce, and the press release goes both directions at once: Salesforce data goes into Claude through a new plugin, and Claude goes into Salesforce as a reasoning model for Agentforce. The release also names Claude "the default model for Slack AI, Slackbot, Salesforce in Claude, Headless 360, Agentforce Coworker, and Claude Code across Salesforce's engineering organization."

The Claudeforce landing page on salesforce.com announcing the Salesforce and Anthropic partnership, as taken from Salesforce
The Claudeforce landing page on salesforce.com announcing the Salesforce and Anthropic partnership, as taken from Salesforce

On September 15, Salesforce announced at Dreamforce that Salesforce in Claude was "now available to all customers in beta," launched alongside AIforce, the layer that exposes Salesforce to any agent through MCP servers, APIs and CLI tools. Anthropic's own write-up, Bringing Salesforce into Claude, says GitLab, Siemens and Legora have deployed it and that 7,000 Salesforce sellers use it.

I build integrations at eesel, so the part I read twice was the roadmap. On the Claudeforce page, Sales is marked Beta. Service is marked "COMING SOON", along with Marketing, Commerce, Field Service and the rest. That one label decides most of what follows for a support team.

The five ways to use Claude with Salesforce

There are five routes, and they split on one question: does Claude reach into Salesforce from outside, or does Salesforce run Claude inside its own product?

Map of the five routes: three where Claude reaches into a Salesforce org and two where Salesforce runs Claude inside
Map of the five routes: three where Claude reaches into a Salesforce org and two where Salesforce runs Claude inside
RouteWhere Claude runsWho signs inStatusBest for
Salesforce in Claude pluginClaude chat and Claude CoworkEach user, with their own Salesforce loginBeta, Salesforce approves orgsSellers: pipeline, renewals, meeting prep
Salesforce (Beta) connector or a custom hosted MCP connectorclaude.ai, Claude Desktop, Claude CodeEach userHeadless 360 server in beta, record servers GAOps and admins who want any object
Claude Tag with the Salesforce connectorSlack channelsOne shared integration userPublic betaTeam channels asking about accounts and deals
Agentforce model optionInside AgentforceSalesforce runs itGA, opt-inTeams already committed to Agentforce
Claude Code with sf-skillsYour terminal or IDEThe developerSkills change oftenApex, LWC and Flow development

If you only remember one thing from the table: the first three are Claude reaching into Salesforce, and none of them was designed around a support queue. The fourth is the only one where Claude acts on customer conversations, and there it is Agentforce doing the acting.

How does the Salesforce in Claude plugin work?

It's a plugin built by Salesforce that bundles 37 sales skills, the Salesforce and Slack connectors, and a first-run skill that learns the user's role and book of business. The skills cover jobs like renewal prep, QBR decks, pipeline coverage and meeting follow-up, and Anthropic's examples of write actions are things like updating opportunities, logging calls and creating follow-up tasks (Anthropic).

Anthropic's announcement page for Salesforce in Claude, dated September 15, 2026, as taken from Claude
Anthropic's announcement page for Salesforce in Claude, dated September 15, 2026, as taken from Claude

Getting it is the hard part. Per the Use Salesforce in Claude help article, it's "available in beta on all paid plans for organizations Salesforce approves through its beta sign-up," and it works in chat and Claude Cowork on web and desktop. On the Salesforce side, your org needs the latest Sales Cloud enterprise edition to be eligible, and you request access through AgentExchange.

Setup is two steps for a Claude Owner. Under Organization settings, Plugins, you pick whether the Salesforce Marketplace plugin is installed by default, available, or required per group. Then under Connectors, you paste an External Client App's consumer key and secret into "Salesforce (Beta)." After that, "each member signs in to Salesforce with their own account the first time they use the plugin."

The safety model is the part I like. Claude "sees only what that user's existing Salesforce permissions already allow," and by default it asks before every write, with Allow once, Always allow or Deny (Anthropic help). The Claudeforce page adds that when Claude updates a record, "it changes only the field it said it would."

So for a sales team on a recent Sales Cloud edition, this is the route. Setup is done once by an admin, it respects your sharing model, and it comes from the vendor that owns your data. For a support team, there's nothing in the box yet.

Connecting Claude to Salesforce with hosted MCP

If you can't get into the beta, or you need objects the sales skills don't touch, Salesforce's hosted MCP servers are the general route. MCP, the Model Context Protocol, is the standard Claude uses to talk to outside tools; Salesforce runs the server, so you don't host anything.

The Salesforce (Beta) connector listing in Claude's marketplace, made by Salesforce and added August 2026, as taken from Claude
The Salesforce (Beta) connector listing in Claude's marketplace, made by Salesforce and added August 2026, as taken from Claude

There are two ways in:

  • The Salesforce (Beta) connector in Claude's directory points at the Headless 360 server, which is in beta and exposes four tools: discover, describe, dispatch and dispatch_readonly. Salesforce says that at launch it offers "dozens of operations, with most focused on Setup tasks for admins" (Headless 360 docs).
  • A custom connector to one of the record servers. The sobject-all server has 11 tools, including SOQL queries, SOSL search, and create, update and delete on any object you have access to, Case included.

Salesforce's Claude setup guide is short. In Claude, go to Customize, Connectors, Add custom connector, paste the server URL, choose "Use your own OAuth client," and paste the consumer key from an External Client App with the mcp_api and refresh_token scopes. Connected Apps aren't supported, and a new app can take up to 30 minutes to start working. One nice detail: "Claude Code inherits the same connection," so you don't add it twice.

Every call runs as the signed-in user, and Salesforce spells out what that means: "If you can't perform an action in Salesforce, your agent can't perform it through the MCP server" (Salesforce). It also recommends setting your client to ask for approval before any tool that changes data.

Hacker News

"My worry would be compliance. On Salesforce projects, security teams approve an integration user with narrow permissions and an audit log much sooner than any kind of browser automation. Can the generated MCP run under a dedicated service account and log every call? Then I could bring it to a client's security review."

That comment is the question I'd expect from any security review. Per-user sign-in answers half of it. The audit trail is still on you.

What does Claude for Salesforce cost?

You pay two vendors, and one of them hasn't published its number yet.

Two receipts: the Claude bill with seats, Tag usage and API tokens, and the Salesforce bill with Flex Credits, MCP calls priced TBA and agent registration
Two receipts: the Claude bill with seats, Tag usage and API tokens, and the Salesforce bill with Flex Credits, MCP calls priced TBA and agent registration

The Claude side is public. Per Claude pricing, checked November 18, 2026 (my Claude pricing breakdown goes deeper):

Claude planPriceNotes
Free$0Connectors available
Pro$17/month billed yearly, $20 monthlyIndividual plan
Max$100/monthIndividual plan
Ultra$200/monthIndividual plan
Team, Standard seat$20 yearly, $25 monthly2 to 150 people, Claude Tag included
Team, Premium seat$100 yearly, $125 monthlyHigher usage
Enterprise$20/seat/month plus usage at API rates20-seat minimum

If you build your own integration on the API instead, the current list prices per million tokens are $2/$10 for Sonnet 5.5, $4/$20 for Opus 5.5 and $0.10/$0.50 for Haiku 5.5 on prompts up to 100k tokens per Anthropic's price list.

The Salesforce side changed in September, and it's the part most Claude for Salesforce guides miss. The hosted MCP docs now say the servers "are intended only for customers with Flex Credits and you may be billed for server usage" (Salesforce). A Salesforce knowledge article from September 17 adds that every successful call a registered agent makes over MCP or the API will be metered as a "Headless Platform Interaction" (KA 005360285).

Three details matter for budgeting:

  • The price per interaction is listed as TBA on the November 3 rate card, and metering isn't on yet. Salesforce promises 30 days' notice.
  • Every agent using a Salesforce MCP server will have to be registered within three months of Salesforce's notice. That applies to all customers.
  • Sandboxes, scratch orgs and Developer Edition orgs aren't metered, so testing stays free.

For reference, Flex Credits cost $500 per 100,000 credits, and a standard Agentforce action is 20 credits, or $0.10 (Agentforce pricing). No price for the Salesforce in Claude plugin has been published; the Claudeforce release says "Pricing and packaging are subject to change."

Claude Tag and Salesforce in Slack channels

Claude Tag is Claude working inside Slack as one shared @Claude, and it has its own Salesforce connector. Anthropic's guide says it lets Claude "read accounts, contacts, opportunities, and cases (and write, if you grant it) in any channel where the connector is on."

The difference from the plugin is identity. Tag connects with one credential, a dedicated integration user, using the OAuth client credentials flow, and Anthropic's advice is to start that user on a read-only permission set. Anyone in a channel where the connector is on can use what that account can see, so the per-service guide says to always connect a dedicated account, never your own login.

One setup gotcha caught my eye: "The first credential you add for a service from the Connectors tab is on in every workspace and channel as soon as you save it." If you have shared or external channels, put the Salesforce credential in a narrower bundle before you save it.

Tag needs Claude Team or Enterprise, bills channel work from a usage balance instead of seats, and isn't available to orgs using zero data retention or the HIPAA setup (Claude docs). Every Salesforce example Anthropic publishes is about deals and accounts, like "which deals are stuck in stage 3." It's a good internal tool for a support lead who wants to ask "what plan is this account on?" without opening Salesforce. My Claude for Slack post covers Tag's limits in customer-facing Slack Connect channels.

Claude inside Agentforce

This is the route that reverses the direction: Salesforce calls Claude, through Amazon Bedrock, inside its own trust boundary.

Salesforce's trust boundary diagram showing Agentforce calling Anthropic through the Salesforce Secure LLM Gateway, as taken from Salesforce Help
Salesforce's trust boundary diagram showing Agentforce calling Anthropic through the Salesforce Secure LLM Gateway, as taken from Salesforce Help

Despite the Claudeforce headlines, Claude is not the default model for Agentforce agents. Salesforce Help still says "We recommend using the Salesforce Default model option," which runs GPT-4.1 for agents built in the new builder. The Claude choice is the AWS-Hosted option, and it "uses Anthropic Claude Haiku 4.5 on Amazon Bedrock" (Salesforce Help). The setting applies to every agent in the org.

The two statements fit together once you read the fine print. The press release says Claude powers Agentforce Vibes and Agentforce Coworker "by default." Those are the developer tool and the employee assistant, not the customer-facing service agent.

If you want a newer Claude model, custom actions are the escape hatch. A prompt template, Apex or the Models API can reference any Salesforce-managed model, and Salesforce now lists 12 Claude models, including Sonnet 5.5, Opus 5.5 and Haiku 5.5. Cost follows the prompt tier on the rate card: Haiku 5.5 bills as a Basic prompt at 2 credits, while Opus models bill as Advanced at 16. Salesforce also warns that Claude "tends to be more sensitive to the nuances in the input payload" than OpenAI models, so plan to retest prompts after switching.

If you're weighing Agentforce itself, my Agentforce review and Agentforce pricing posts go deeper.

Claude Code for Salesforce developers

For developers, the biggest change is a quiet one. Salesforce's DX MCP server, the @salesforce/mcp package that many Claude Code tutorials still point to, was archived on November 2, 2026. The GitHub repo is now read-only, and its data toolset only ever held one read tool, run_soql_query.

The replacement is skills. Salesforce now publishes forcedotcom/sf-skills, covering Agentforce agents, Lightning apps, Flow, Apex, SOQL, LWC and permission sets, plus a Salesforce Development plugin for Claude Code that the AIforce announcement says brings "more than 40 skills." The repo carries a fair warning: "Expect frequent changes."

Claude Code building and deploying a Salesforce Lightning app with the Salesforce Development plugin, as taken from forcedotcom/sf-skills on GitHub
Claude Code building and deploying a Salesforce Lightning app with the Salesforce Development plugin, as taken from forcedotcom/sf-skills on GitHub

For data access from Claude Code, use a hosted MCP server; Salesforce's developer blog walks through it, and once the connector is in claude.ai, Claude Code picks it up. Agentforce Vibes, Salesforce's own coding agent, runs on Claude and Mastra orchestration engines, so you're often using Claude either way. My best AI for Salesforce coding roundup compares the options.

LinkedIn

"The whole thing got built in one Claude Code session. Apex controller, the LWC bundle, the modal, all of it. Then deployed straight to my dev org through the Salesforce DX MCP server."

That post is from June, before the DX server was archived. The workflow still works; the plumbing underneath it moved to skills.

Where Claude for Salesforce stops for support teams

Here's where I'd slow down if you run a Service Cloud queue. Each route gets a support case part of the way, and none of them takes it to the customer.

A four-step staircase from reading a case to sending the reply, showing how far each Claude route climbs before a support agent is needed
A four-step staircase from reading a case to sending the reply, showing how far each Claude route climbs before a support agent is needed
  • Salesforce in Claude ships sales skills only. Service skills are "COMING SOON" on the Claudeforce page, with "Auto setup agents from cases" and "Optimize contact center operations" listed, and no date.
  • Hosted MCP can read a Case and update its fields, but there is no purpose-built reply-to-case or send-email tool in the standard catalog. A reply would mean creating records through the generic tools, with a person approving each one.
  • Claude Tag answers your team in Slack. It's an internal assistant, not a responder on the case.
  • Agentforce with Claude does answer customers, but you're buying Agentforce, with Claude Haiku 4.5 as the model and Salesforce's per-action or per-conversation pricing on top.
LinkedIn

"Agentforce can automate customer-facing tasks, such as talking to customers and handling their issues. Claudeforce is primarily an internal AI assistant that helps the sales team find information and automate their own tasks."

That split matches what I found in the docs. If your goal is fewer cases waiting on a person, read my deeper Claude for Salesforce Service Cloud guide, which walks through what each route does on a Case, and the wider AI for Salesforce Service Cloud roundup.

Which route should you pick?

Pick by team, not by model.

You areUseWhy
A sales team on the latest Sales Cloud enterprise editionSalesforce in Claude pluginVendor-built skills, per-user permissions, approval on every write
An admin or ops lead who needs any objectHosted MCP connectorFull record access as yourself, test free in a sandbox
A team that lives in SlackClaude Tag with a read-only integration userAnswers account questions in channels without seats
Already on Agentforce, want Claude's reasoningAWS-Hosted option or custom actionsStays inside Salesforce's trust boundary
A Salesforce developerClaude Code with sf-skillsThe DX MCP server is archived
A support team that wants cases handledA support teammate on the case, like eeselNone of the Claude routes reply to customers yet

The trap I'd avoid is wiring a homemade support bot onto hosted MCP right now. You'd be building reply logic Salesforce doesn't provide, on calls that will soon carry a meter with no published price. Wait for the numbers, or use something built for the queue.

How eesel handles Salesforce cases

I'll be upfront about how eesel fits: it doesn't replace Claude for your sellers. It's the support teammate for the case queue, and it plays nicely with Claude.

The eesel AI for Salesforce page describing AI agents that connect to Service Cloud and work through tier 1 cases, as taken from eesel
The eesel AI for Salesforce page describing AI agents that connect to Service Cloud and work through tier 1 cases, as taken from eesel

Per the eesel Salesforce docs, the agent "works inside your Salesforce org. It reads cases and Knowledge articles (and contacts and accounts once you switch them on), drafts replies as internal case comments, and updates records." It can also send the email to the customer, close a case, or update a contact or opportunity, and each action can require approval, which is how I'd set agent permissions on day one (actions and approvals). Setup is a managed package a Salesforce admin installs, and the agent "only sees what the Salesforce user you authorized can see."

That also answers the security reviewer's question from earlier. Each activity entry links to the Salesforce record and shows what eesel searched, each action it took, who approved it and why.

The Claude angle is the part I like most. Once Salesforce is connected, you can talk to your eesel agent from Claude or ChatGPT or Cursor, and it works with your Salesforce data. So a support lead can ask Claude "what's open for this account?" and get the answer from the same teammate that's working the queue.

What this looks like in practice: at MentorShow, eesel reads each member's subscription status, renewal date and past tickets from the CRM record before drafting, and refunds and billing go straight to a person. It read 1,200 conversations and tickets and wrote 432 reply drafts for the team in a month. That's HubSpot rather than Salesforce, but it's the same job: the CRM record changes the answer. On Salesforce itself, CartonCloud's service desk lead said eesel keeps "that human touch" while getting agents to the right articles quickly (eesel pricing; CartonCloud case study).

Where eesel isn't the right fit: if you want Claude to prep sellers for renewals, the Salesforce in Claude plugin does that, and eesel doesn't try to. eesel also has no trigger for Salesforce chat or messaging sessions; it works on records like Cases.

If you'd rather set it up from a terminal, the eesel CLI lets Claude Code run eesel integrations connect salesforce, download your knowledge sources and switch on automations, then ask the agent test questions before anything goes live. The Salesforce connection itself still needs an admin to approve it in the browser.

Try eesel on your Salesforce cases

Claude for Salesforce is a strong sales tool today and a promising support tool later. If your cases can't wait for "coming soon," eesel for Salesforce joins your Service Cloud queue, learns from past cases and Knowledge articles, drafts or sends replies with the approvals you set, and answers from Claude when your team asks. Plans start at $299 a month for 500 credits (eesel pricing), so you can try eesel on a slice of your queue before you commit.

Frequently Asked Questions

Is there an official Claude for Salesforce integration?
Yes. Since September 15, 2026, Salesforce publishes Salesforce in Claude, a beta plugin with 37 sales skills, plus a Salesforce (Beta) connector in Claude's directory. Both run on Salesforce's hosted MCP servers. For the support side, the Claude for Salesforce Service Cloud guide goes route by route.
How much does Claude for Salesforce cost?
There are two bills. Claude Team seats run $20 to $125 a month depending on seat type and billing, per Claude pricing. On the Salesforce side, hosted MCP servers are meant for orgs on Flex Credits, and agent calls will be metered as Headless Platform Interactions once Salesforce sets a price. No price for the Salesforce in Claude plugin itself has been published.
Can Claude reply to Salesforce cases for my support team?
Not out of the box. The shipped Salesforce in Claude skills are for sellers, and service skills are listed as coming soon on the Claudeforce page. Over hosted MCP, Claude can read and update a Case with your approval, but there is no reply-to-case tool. A support teammate like eesel for Salesforce is built for that job.
Is Claude the default model in Agentforce?
No. Salesforce Help still recommends the Salesforce Default option, which runs GPT-4.1 for agents built in the new builder. Claude Haiku 4.5 is the opt-in AWS-Hosted option. Claude is the default inside Agentforce Vibes, Coworker and Slackbot, which is where the Claudeforce headlines come from.
Which Claude plan do I need to connect Salesforce?
The Salesforce in Claude plugin works on all paid Claude plans, but only for organizations Salesforce approves and only with the latest Sales Cloud enterprise edition. Claude Tag, which brings Salesforce into Slack channels, needs Claude Team or Enterprise. A custom connector to a hosted MCP server works from any plan with connectors.
Can I use Claude Code to build on Salesforce?
Yes. Salesforce archived its DX MCP server on November 2, 2026 and now ships tooling as skills in forcedotcom/sf-skills, including a Salesforce Development plugin for Claude Code. If you add a hosted MCP server as a connector in Claude, Claude Code inherits it.

Share this article