← Insights

Salesforce Agentforce Billing: Model by Workflow, Not Per Action

Flex Credits now meter every agentic MCP call at $0.10 list. One 'conversation' can chain dozens of actions — your budget model needs to change before your rollout does.

The Cost Model Salesforce Operators Are Missing

Salesforce has confirmed that Flex Credits will meter agentic MCP calls — 20 credits ($0.10 at list price) per Agentforce action or Data Cloud 360 query triggered via MCP. Standard record reads still consume existing API allocations. But once you cross into the agentic layer, every tool call has a price tag.

The trap is thinking in per-action terms. A single user-facing 'conversation' — say, a rep asking an agent to pull an account summary, check open cases, and draft a follow-up — can chain four, six, or eight discrete tool calls behind the scenes. What looked like a $2 flat interaction under old pricing models now has a variable tail. If you're planning workflow volume at any scale, that tail matters.

Two additional constraints make this urgent:

  • Flex Credits expire at Order End Date with no rollover. Overbuy for a rollout that slips, and you lose the buffer.
  • Salesforce promises only 30 days' notice before currently unmetered usage types flip to metered. Hosted MCP record operations are explicitly called out as a candidate. Thirty days is not enough runway to re-architect a workflow you've already deployed to the field.

What Operators Need to Do Before They Commit

The CFO conversation has changed. It's no longer 'what does an Agentforce seat cost?' It's 'what is the average tool-call chain depth for each workflow we plan to run, and what does that look like at P50 and P95 volume?'

That requires three things most teams haven't built yet:

  1. A workflow inventory that maps to agentic touchpoints. Know which processes touch Agentforce, Prompt Builder, or Data Cloud queries — and which don't. The metered boundary is the thing to audit, not the license count.
  2. Chain-depth estimates per workflow. Work with whoever is building your agent flows to instrument or estimate how many tool calls a representative interaction generates. Even rough numbers let you build a defensible range.
  3. A trigger for the 30-day notice. Assign someone to watch Salesforce's metering change communications. If Hosted MCP flips to metered and you haven't modeled it, you're absorbing an unbudgeted cost mid-fiscal year.

The Broader Pattern

This is the first clear breakdown of how MCP plus agentic architecture translates to real spend — and it previews what every enterprise AI platform will eventually do. Agentic systems are inherently multi-step; that's the value. But multi-step means multi-event billing, and that's a fundamentally different cost structure than per-seat or per-query SaaS.

Operators who do the workflow-volume modeling now will negotiate better, deploy more deliberately, and avoid the mid-year variance conversation. The ones who skip it are setting up a budget surprise timed to their first real agentic rollout.