Salesforce will charge Flex Credits for every successful MCP or direct API call made by a registered AI agent, metered as a "Headless Platform Interaction" (HPI) in Digital Wallet. The multiplier is still "TBA," nothing is metered yet, and Salesforce promises 30 days' notice before metering begins. Only active production orgs are metered, and traditional integrations keep current pricing. For RIAs, banks, and insurers piloting Claude or other agents on Salesforce, the job now is to inventory production agents, separate agent identities from integration users, and baseline call volumes. Vantage Point does this work as part of Salesforce architecture and governance engagements.
Every Claude-to-Salesforce MCP call in production is about to have a price tag. Salesforce just hasn't said what it is.
In a Help article published September 17, Salesforce laid out a new billing model for agentic access to its platform: once an AI agent is registered, every successful call it makes into Salesforce is counted as a Headless Platform Interaction and draws down Flex Credits. Salesforce Ben reported the change on September 22 — the same day Anthropic released Claude Opus 5.5 and OpenAI launched GPT-6 Sol and Luna, both at lower API prices. The model got cheaper the day the platform call got a meter.
For regulated firms that spent Dreamforce week scoping Claude-plus-Salesforce pilots, the questions are practical: what is metered, who is exposed, and what should you do before Salesforce publishes the number?
Salesforce is tying who an agent is to what it costs. According to the Help article, it is targeting November to release a new set of security controls, agent registration, and a new billing model as part of its Headless Toolkit — the functionality behind AIforce, which was announced at TDX 2026 as "Headless 360." In Salesforce's words, "the same registration step that gives an external agent its own identity on the Platform is what makes that agent's activity visible, governable, and billable."
Salesforce Ben notes that customers bringing Claude or ChatGPT into Salesforce already pay those providers; Flex Credits add a second cost on the Salesforce side. As SF Ben's Tim Combridge put it, "Organizations will need to know what they're looking to achieve with their agents, and estimate the cost of running them before they can make a decision on whether or not to proceed."
How the published rules apply, per Salesforce's Help article:
| Situation | What Salesforce says |
|---|---|
| Registered agent calling Salesforce over MCP in a production org | Metered as an HPI once the multiplier is published and notice is given |
| Registered agent calling a Salesforce API directly in production | Metered as an HPI — the model covers agentic API traffic, not just MCP |
| The same agent in a sandbox, scratch org, or Developer Edition org | Not metered through Digital Wallet |
| Traditional, non-agentic integration (for example, a scheduled middleware sync) | Current pricing and security options unchanged |
| Any use of a Salesforce MCP server | Agent registration required within three months of Salesforce's notice that Agentic Identity is available — for all customers |
| Existing customers with agents using Salesforce APIs | Register all agents and migrate connections to the new billing model upon renewal |
| Customers purchasing on or after September 17, 2026 | Register all agents that use Salesforce APIs within three months of the Agentic Identity notice |
At RIAs, banks, and insurers, the exposed workloads are the ones that looked best at Dreamforce: Claude prepping an advisor for a household review, an agent summarizing service cases, an assistant drafting renewal outreach from policy data. Each touches Salesforce through MCP or an API, as a registered agent.
The edges are less clear. Salesforce hasn't published a test for where "agentic" traffic ends and "traditional integration" begins — say, a middleware flow kicked off by an agent. Get your account executive's answer on borderline workflows in writing.
Building agents is free. Running them isn't. Because sandboxes, scratch orgs, and Developer Edition orgs aren't metered, a pilot can run for months without a single metered call. Salesforce adds that credits seeded into the org cover deploying to production and testing there. The effect: most teams first see an agent's unit cost after real users hit production.
Agents also don't use Salesforce the way people do. An agent answering one question may search, retrieve several related records, and write a note — each a separate call. Until you measure it, you don't know how many calls one business outcome takes. A pilot business case that shows the Salesforce side of agent cost as zero should show "TBA" instead.
None of these three moves depends on the multiplier.
You can't register, govern, or budget for agents you can't see. List everything that calls your production org on behalf of an AI: Claude or ChatGPT connectors using MCP, internally built agents, vendor tools with embedded agents, and scripts someone wired to an API token. For each, record the owner, connected app, OAuth flow, the user it authenticates as, the objects it touches, and the workflow it supports. Connected app usage, login history, and Event Monitoring (where licensed) are good starting points.
Salesforce's resolution steps describe reconfiguring MCP clients and API-based agents to present OAuth credentials tied to the registered agent — "the step that redefines the connection and allows Salesforce to classify the connecting app as an agent." The logical implication: how each connection is classified is what determines which calls are metered. Today, many agents run under a shared integration user or a human's login, blurring agent and system traffic.
Untangle that before registration opens. Give each agent its own connected app and scoped permission set, keep true system-to-system integrations on dedicated integration users, and retire any agent riding a person's credentials. The goal is accurate classification, not avoidance: passing agent traffic off as integration traffic is a contract risk and, for a regulated firm, a supervision problem. The upside: a distinct, least-privilege identity per agent gives compliance a clean answer to "which agent did what, with what access."
Measure how many calls each agent makes per unit of business work — per meeting prep, case summary, or renewal draft — over representative weeks. For pilots still in a sandbox, instrument test runs to count calls per task and pair that with expected production volumes. When the multiplier is published, the cost model is simple arithmetic: calls per outcome, times outcomes, times the multiplier, times your contracted credit rate. Do that math the day the rate card changes, not on the first statement.
For how Flex Credits and Digital Wallet work today, see our Salesforce Flex Credits pricing and usage guide.
| Item | Status | Source |
|---|---|---|
| Billing model documented | Published September 17, 2026 | Salesforce Help |
| HPI usage type on rate card | Listed under Customer 360 Platform | Flex Credits Rate Card |
| HPI multiplier (credits per call) | TBA | Flex Credits Rate Card |
| Security controls, agent registration, billing release | Salesforce is targeting November | Salesforce Help |
| Metering start | Not before 30 days' notice after a multiplier is added | Salesforce Help |
| Agentic Identity availability date | Not announced | — |
| MCP server registration deadline | Three months after Salesforce's Agentic Identity notice | Salesforce Help |
| Existing customers' API-based agents | Register and migrate at renewal | Salesforce Help |
Vantage Point helps regulated firms reach production with governance and a cost model already in place. Through our Salesforce implementation and advisory services, we run the agent and MCP inventory, separate agent and integration identities, and instrument pilots for calls-per-outcome baselines. Our compliance and security solutions team scopes agent permissions and audit trails, and managed services keep consumption reviews running after go-live. For the broader architecture, see our Headless 360 architecture and readiness guide.
In one Vantage Point engagement, an insurance brokerage took Agentforce into production (Phase 4) with 100% user login at go-live, at the end of a 56-week transformation. Senior consultants only — no junior handoffs; the experts you meet are the experts who deliver.
The inventory, identity, and baselining work doesn't need the multiplier. Vantage Point's senior consultants can map every agent touching your production org and build a calls-per-workflow baseline so you're ready the day the rate card changes. Contact Vantage Point to schedule an agent metering readiness review, or explore our Salesforce services.
A Headless Platform Interaction (HPI) is Salesforce's new Flex Credits usage type for calls that registered AI agents make to the Salesforce platform. Every successful call a registered agent makes, whether over MCP or direct API, counts as one HPI and is tracked in Digital Wallet.
Nobody knows yet: the Flex Credits Rate Card lists the Headless Platform Interaction multiplier as "TBA." Salesforce's pricing page lists Flex Credits at $500 per 100,000 credits, but without the multiplier that can't be converted into a per-call cost, so treat any per-call estimate as speculation.
Not before Salesforce publishes a multiplier and gives 30 days' notice. Agentic usage isn't metered today. Salesforce is targeting November to release agent registration, security controls, and the billing model, but has not announced when metering begins.
No. Salesforce says Headless Platform Interactions are metered only in active production orgs. Sandboxes, scratch orgs, and Developer Edition orgs aren't metered, so teams can build and test agents without drawing down Flex Credits — which is also why unit costs surface at go-live.
No. Salesforce says the pricing and security options for traditional integrations stay exactly as they are today, and the new model applies specifically to agentic traffic from registered agents. Confirm borderline workflows with your account executive in writing.
Agentic Identity is Salesforce's upcoming agent registration feature. Each agent gets its own identity and scoped permission set instead of operating under a human user's credentials. MCP clients and API-based agents then connect with OAuth credentials tied to the registered agent, which is how Salesforce classifies the connection as agentic.
Yes. Salesforce says use of any Salesforce MCP server requires agent registration within three months of its notice that Agentic Identity is available, for all customers. Existing customers with API-based agents must register them and migrate at renewal.
Inventory every agent and MCP client touching production, then separate agent identities from integration users and baseline calls per workflow. Vantage Point runs this readiness work for RIAs, banks, and insurers as part of Salesforce advisory and governance engagements.
Vantage Point is a boutique CRM consulting firm helping businesses transform with Salesforce, HubSpot, and AI — 150+ clients, 400+ engagements, and a 4.71/5 average engagement rating. Learn more at vantagepoint.io.