Salesforce Will Meter AI Agent API and MCP Calls in Flex Credits: What It Means for Teams Connecting Claude, ChatGPT or Their Own Agents to a CRM

Salesforce is going to charge for the work AI agents do inside its platform. If you connect Claude, ChatGPT, Codex or an agent your own team built to a Salesforce production org, each successful call that agent makes will be counted as a Headless Platform Interaction and paid for in Flex Credits once metering starts. How many credits one call costs has not been published.

Below: what Salesforce has said, quoted from its Help article and rate card, what is still unknown, how to size the risk, HubSpot’s October changes, and what to ask any CRM vendor. Facts are current as of November 13, 2026; check the sources at the end before you budget.

What Salesforce announced

The source is a Salesforce Help article titled “Understand AIforce Impact” (Knowledge Article 005360285), published September 17, 2026. It explains that AIforce, “originally announced at TDX 2026 as ‘Headless 360,'” lets agents work with Salesforce outside the browser, and that the Headless Toolkit brings “new license and billing considerations.”

The article says that, “Targeting November, Salesforce will be releasing a new set of security controls, agentic registration, and a new billing model as part of the Headless Toolkit.” It ties the pieces together: the step that gives an external agent its own identity on the platform is the same step that makes its activity “visible, governable, and billable.”

Three parts matter most for teams that connect AI to a CRM:

• Agent registration and Agentic Identity. Admins register each agent on the platform. Registration gives the agent its own identity “instead of letting it operate under the identity of the person it assists,” and admins can give that identity a narrower permission set than a human user.
• A new usage type. A Headless Platform Interaction usage type has been added to the Customer 360 Platform category of the Flex Credits rate card. In the article’s words, “Every successful call a registered agent makes to Salesforce, whether it arrives over Model Context Protocol (MCP) or direct API, is metered as a Headless Platform Interaction (HPI) and tracked through Digital Wallet.”
• A registration deadline. Customers buying on or after the publication date must register all agents that use Salesforce APIs within three months of notice that Agentic Identity is available. Existing customers using agents with APIs must register them and move their connections to the new billing model “upon renewal.” Use of any Salesforce MCP server requires registration within three months of notice, for all customers.

For the full background on Salesforce’s own MCP tooling, see our guide to the Salesforce MCP server.

What counts as a metered call, and what does not

Metered, once the multiplier is published

• Successful calls from a registered agent to Salesforce, whether over MCP or a direct API call.
• Only in active production orgs. The article says Flex Credits “will only be consumed for HPIs occurring in active production orgs.”
• Tracked in Digital Wallet, the same place Salesforce shows other Flex Credit usage.

Not metered

• Sandboxes, scratch orgs and Developer Edition orgs. The article says they “don’t incur HPI metering through Digital Wallet, so admins and developers can build and test without drawing down Flex Credits.”
• Traditional integrations. “The pricing and security options for traditional integrations stay exactly as they are today.” The model “addresses agentic traffic specifically.”
• Anything before the multiplier is set. The usage type is listed as “TBA” and “agentic usage isn’t metered” until Salesforce adds a number and gives notice.

Two points are easy to misread. First, metering attaches to registered agents, but registration is not optional for long: the same article makes it a requirement within three months of notice, or at renewal for existing API customers. Second, the article does not define where an “agent” ends and a “traditional integration” begins. If a model drives one of your sync jobs, ask your account executive how it will be classified.

The rate card adds one more detail in a footnote: the Headless 360 usage types “are in addition to other Usage Types” on the card. An agent that also triggers other metered features may consume credits under more than one line.

Why the price is still unknown

Flex Credits themselves have a list price. Salesforce’s Agentforce pricing page shows Flex Credits at $500 per 100,000 credits, with a note that the page “is provided for information purposes only and is subject to change.” That works out to half a cent per credit at list.

What is missing is the multiplier: how many credits one Headless Platform Interaction consumes. The Flex Credits rate card dated November 3, 2026 lists the Customer 360 Platform usage type “Headless Platform Interaction” with a multiplier of “TBA.” The Help article confirms it: “The multipliers for this usage type haven’t been announced yet.” Salesforce says it “will provide 30 days’ notice when a numerical multiplier is added and metering begins.”

So any per-call dollar figure quoted today is a guess. The same rate card lists 20 credits for a standard Agentforce action in production, but that is a different usage type; do not assume an agent API call will be priced the same way.

The article does not say how retries are treated (only “successful” calls are metered), whether high volumes get tiered rates, or how negotiated contracts will price HPIs. Ask your account executive in writing.

How to estimate your exposure

You cannot calculate a bill yet, but you can calculate your call volume, which is the number the multiplier will be applied to. Work it out in four steps.

1. List every agent that touches the CRM. Include assistants like Claude or ChatGPT that reps connect through MCP, agents your developers built, and agents inside other tools that call Salesforce for you.
2. Count tasks per agent per day. For assistants, that is roughly the number of CRM requests each rep makes per day, multiplied by the number of reps using it. For background agents, it is the number of records or events they process.
3. Count calls per task. One request in plain language often becomes several API calls: a search, a read of the record, a read of related records, then a write. Every successful call is its own HPI. Run a typical task in a sandbox, where nothing is metered, and count the calls in your logs.
4. Multiply by working days. Agents per team, times tasks per agent per day, times calls per task, times working days in the month, gives monthly HPIs.

When Salesforce publishes the multiplier, monthly HPIs times the multiplier gives credits, and credits times your contract price per credit (half a cent at list) gives dollars. Until then, keep a range for a quiet month and a busy month. Agents that loop, poll or re-read the same record are where costs will concentrate.

What HubSpot changed in October

Salesforce is not the only CRM moving AI into usage-based billing. CX Today reported on October 20, 2026 that HubSpot “has introduced more flexible pricing options designed to give businesses greater control over spending on AI-powered tools,” including AI spend caps, the ability to exchange seats for credits, and flexible payment options.

HubSpot’s own pricing page describes HubSpot Credits as the way to use its AI agents and features: credits are used “when features take action across the platform,” you can buy them in packs or pay as you go, and you can “adjust default spend controls to match your budget and set action-level credit limits.”

Caps make spending easier to control, but a capped feature can stop when the cap is reached. CX Today framed the buyer’s question as whether these controls make costs easier to forecast as adoption grows. For seat costs, see HubSpot per-seat pricing.

Questions to ask any CRM vendor about agent pricing

Get written answers to these before you connect agents at scale or renew:

1. Is a call from an AI agent priced differently from one by a person or a traditional integration?
2. What is the unit, and where is the published rate?
3. Do MCP calls and REST API calls cost the same?
4. Are failed calls, retries and reads metered, or only successful writes?
5. What happens when credits or a cap run out? Does the agent stop or keep running at an overage rate?
6. Are test environments free? Salesforce says sandboxes are not metered; confirm the same elsewhere.
7. How much notice do you give before a rate changes, and does a change apply mid-contract or at renewal?
8. Can each agent have its own identity and a narrower permission set, and can you see which agent made which call?
9. Is there a cap on the number of agents you can connect, separate from the number of human users?

Our flat-rate CRM pricing guide has a wider list of questions to ask before signing, and AI CRM without credits explains how credit-based AI pricing works in general.

How Conduyt handles agent and API calls

Conduyt is a CRM priced as a flat monthly rate. The pricing page lists Starter at $299 a month and Professional at $499 a month, each with unlimited users, unlimited contacts and unlimited pipelines, and states “No per-seat fees. No usage metering.” There are no credits for AI or API calls.

• A public MCP server with 390+ tools. The Conduyt MCP server works with Claude Desktop, Claude Code, ChatGPT and custom agents that speak Model Context Protocol, and is included on every plan. The page puts it plainly: “No per-seat fees. No API metering. No premium AI add-ons.”
• The REST API behind it. Conduyt publishes 980+ REST API operations across 600+ endpoints, and API access is on every paid plan and in the trial.
• Agent controls without a meter. MCP tool calls are authenticated with your API key, scoped to your account’s permissions and logged, and tokens can carry scopes and action budgets that bound what an agent can do.
• AI teammates in Team Chat. Team Chat puts your people and the Claude Code, Codex or Grok agents they already run into the same channels inside the CRM. Each AI teammate acts with its owner’s Conduyt permissions, and anything it drafts to send in your name waits for Approve, Edit first or Decline. AI teammates are switched on in every 20-day trial and included on Professional. Setup steps are in Connect your AI to Team Chat, and you can read a real thread of three agents finishing a task in how AI agents work together in a CRM.

If you are weighing a move, compare the two directly in Conduyt vs Salesforce, or see the wider field in Salesforce competitors and alternatives. The competitors guide also says where Salesforce still fits better, such as multi-cloud orchestration and industry clouds.

FAQ

Will Salesforce charge for MCP calls from Claude or ChatGPT?

Yes, once metering starts. Salesforce’s Help article says every successful call a registered agent makes over MCP or direct API is metered as a Headless Platform Interaction and consumes Flex Credits in production orgs. It does not matter which AI model is behind the agent.

How much will a Salesforce agent API call cost?

Nobody outside Salesforce knows yet. The Flex Credits rate card lists the Headless Platform Interaction multiplier as “TBA.” Flex Credits list at $500 per 100,000, but without the multiplier you cannot turn a call into a dollar amount.

When does Salesforce start metering agent calls?

Salesforce said it was targeting November 2026 for the registration and billing changes, and that it will give 30 days’ notice before a numerical multiplier is added and metering begins. Until then, agentic usage is not metered.

Are Salesforce sandbox calls charged?

No. The Help article says sandboxes, scratch orgs and Developer Edition orgs do not incur Headless Platform Interaction metering, so teams can build and test agents without drawing down Flex Credits.

Can I avoid metering by not registering my agents?

Not for long. Metering applies to registered agents, but Salesforce requires registration: new customers within three months of notice that Agentic Identity is available, existing customers using agents with APIs at renewal, and anyone using a Salesforce MCP server within three months of notice.

Do traditional Salesforce integrations change?

Salesforce says the pricing and security options for traditional integrations stay as they are today and that the new model addresses agentic traffic specifically. Confirm with your account executive where the line falls.

Does Conduyt charge for AI agent or API calls?

No. Conduyt plans are a flat monthly rate with unlimited users and no usage metering. The MCP server with 390+ tools and API access are included on every plan, and AI teammates in Team Chat are on in every 20-day trial.

Sources

Checked November 13, 2026. Pricing and dates change, so confirm with the source before relying on them.

• Salesforce Help: Understand AIforce Impact (Knowledge Article 005360285, published September 17, 2026: HPI definition, November target, 30 days’ notice, registration, sandboxes, traditional integrations)
• Salesforce: Flex Credits Rate Card, November 3, 2026 (Headless Platform Interaction multiplier listed as TBA; other usage type multipliers)
• Salesforce: Agentforce pricing (Flex Credits at $500 per 100,000 credits, information purposes only and subject to change)
• CX Today: CRM Vendors Tackle AI Costs and Customer Data Access (October 20, 2026: HubSpot AI spend caps, seats exchangeable for credits)
• HubSpot: pricing (HubSpot Credits, packs or pay as you go, spend controls and action-level credit limits)

Related: AI CRM without credits, flat-rate CRM pricing and Salesforce MCP server