Claudeforce: what Salesforce and Anthropic actually ship

Updated October 21, 2026 · Published October 21, 2026

Claudeforce is the partnership Salesforce and Anthropic announced on 26 August 2026: Claude becomes a reasoning model inside Salesforce, and Salesforce becomes a plugin inside Claude, with 37 ready-made sales skills. The plugin is in beta for approved Enterprise orgs, and it still has no published price.

This guide sticks to what is published and dated. Where Salesforce or Anthropic have not published something, it says so.

What Claudeforce is, in two directions

The announcement covers two separate things that are easy to mix up.

Claude inside Salesforce. Claude is available as a reasoning model in Agentforce, including its Atlas reasoning engine, and is the default model in Agentforce Vibes and Agentforce Coworker. In Slack, Claude is the default model. This part was live on the day of the announcement.

Salesforce inside Claude. A plugin named Salesforce in Claude brings a seller's accounts, opportunities and pipeline into Claude, under that seller's existing Salesforce permissions. It ships with 37 skills, a Salesforce connector and a Slack connector. Anthropic released it in beta on 15 September 2026.

The second direction is the one that changes daily work for sales teams, and the one most people mean when they search for Claudeforce.

The names, in order

Claudeforce sits on a platform layer that has changed name three times in six months. The dates matter, because guides written in May and October describe the same thing under different names.

What the 37 skills do

Salesforce does not list all 37 skills on its product page. Anthropic and Salesforce describe them by job:

• Morning brief: the day's meetings, deals closing soon, at-risk opportunities and unread threads, with actions such as pushing a close date.
• Call prep: open opportunities, the week's account team discussion and open questions, from Salesforce, Slack and email.
• Deal review: scores an opportunity against the team's methodology, drafts a business case and a dated mutual close plan, and adds missing stakeholders. Nothing saves until the seller approves.
• Post-call update: turns a conversation into CRM updates and follow-up tasks.
• Account growth: account planning, lead triage and customer health.

All of these are seller-in-the-loop: a person asks, reads and confirms. The plugin is not an unattended job that reviews the whole pipeline every night.

Who can use it today

To join the beta, an org needs:

1. the latest Sales Cloud Enterprise edition;
2. a paid Claude plan;
3. Salesforce's approval, requested by a Salesforce admin through the beta sign-up on AgentExchange.

Setup is one org-level connection. The Salesforce admin turns on the Salesforce MCP server and creates an External Client App; the Claude admin enters its consumer key and secret under Organization settings, Connectors, "Salesforce (Beta)". Each member then signs in with their own Salesforce account on first use, so every call runs with that person's permissions.

What it costs, and why nobody can quote it yet

There is no list price. A seller's action is metered twice, by two vendors, on two contracts.

Salesforce's president put it plainly to VentureBeat, quoted by Beri: "You can't buy this on one piece of paper at the moment." To budget the Salesforce side, the closest published reference is Agentforce's own price list, covered in Agentforce pricing.

How to prepare your org

Cloud Odyssey's beta guide sums up the preparation in three moves, and each one is cheaper before the plugin is live than after.

1. Audit permissions. The plugin acts as each seller, so whatever a seller can see or edit, Claude can too. Review profiles, permission sets and sharing rules on opportunities, accounts and contacts, and remove broad edit rights that nobody needs.
2. Clean the data the skills read. Close dates in the past, empty next steps, deals with no activity for weeks: the morning brief and the deal review will repeat them as facts. A one-time cleanup of open opportunities in the current quarter is the highest-return task before the beta.
3. Pilot in a sandbox, then with a small team. Agent metering planned for November 2026 excludes sandboxes, and a first run that writes to records belongs there. Then give the plugin to five sellers for a month and count what they actually use.

Two further checks are worth adding: confirm that your org is on the latest Enterprise edition before applying, and decide who reads the API allocation each week, since the plugin's calls share it with every other integration.

What Claudeforce does not change

• The data still has to be right. Every skill reasons on your opportunities, close dates and activities. A deal that has been silent for 90 days but still sits in Commit is summarized as faithfully as a real one.
• Someone still has to ask. The plugin answers the seller who opens Claude. A manager who wants every rep's pipeline checked before Monday's review still needs a report, a dashboard or a nightly job.
• Other editions wait. Orgs on editions below the latest Enterprise edition are outside the beta for now.

For teams that want the unattended part, the manager's view every morning at a fixed price, that is what TruthPipe does. The reasoning model underneath Agentforce is also changing: Salesforce's own model, Koa, is in pilot.