Every Salesforce integration review now starts with one question: is this traffic agentic? Salesforce has said that successful MCP and API calls from registered AI agents will consume Flex Credits, while traditional integrations keep their current pricing. That makes the line between "agent" and "integration" a design, security and budget decision.
For the news itself, dates and what's still "TBA," see our post on how Salesforce will meter agentic MCP and API calls. This guide covers the next step: how to classify each connection into your org, what to change in identity design, and what to ask vendors before renewal.
Status as of September 28, 2026: Salesforce's Flex Credits Rate Card still lists the Headless Platform Interaction multiplier as "TBA," and agentic usage is not metered yet.
Agentic Salesforce traffic is API or MCP traffic where an AI agent decides at runtime which calls to make, and that agent connects with its own registered Agentic Identity. Traffic from a designed integration, where code, a flow or a mapping fixes the sequence, is non-agentic, and Salesforce says its pricing stays as it is today. To classify a connection, ask two questions: who decides the next call, and whose credentials does it present? Integration architects, admins and RevOps leaders should run that test on every connection before Agentic Identity arrives. Vantage Point runs these integration reviews as part of Salesforce architecture work.
Salesforce hasn't published a formal test, but its own documents give you two signals to work with.
Signal 1: who decides the sequence. Salesforce's architecture guidance on how Headless 360 changes integration architecture draws the line clearly. "In a traditional integration, you design the sequence," it says. With agents, "the agent determines the execution order at runtime based on its own reasoning." The patterns stay the same. What's new is "a non-deterministic caller."
Signal 2: whose credentials the call presents. Salesforce's Help article, Understand AIforce Impact, says admins will register each agent so it gets "a discrete identity" rather than operating as the person it assists. MCP clients and API-based agents then present OAuth credentials tied to the registered agent. That is "the step that redefines the connection and allows Salesforce to classify the connecting app as an agent." A Salesforce spokesperson told Salesforce Ben that registered third-party agent traffic is identified and isolated separately from traditional API activity.
In practice, Signal 1 tells you what a connection is, and Signal 2 is how Salesforce will see it. Your job is to make them match.
Run every connection into your production org through these questions, in order:
This table applies the two signals to connection types most orgs have. It's a starting point for review, not a Salesforce ruling.
| Connection type | Likely classification | What to check |
|---|---|---|
| Middleware flows (MuleSoft, Workato, Boomi) on schedules or events | Traditional integration | Whether any flow is exposed as a tool that an AI agent calls |
| Middleware exposing Salesforce to an AI agent as tools | Gray zone, leaning agentic | Who decides the calls: the agent upstream or a fixed flow |
| ETL, data warehouse and backup tools | Traditional integration | Whether the vendor has added AI features that call Salesforce |
| Managed-package and ISV apps | Depends on the feature | Whether an embedded agent calls Salesforce APIs from outside the org |
| Claude, ChatGPT or other MCP clients | Agentic | Which MCP server they use and whose login they use today |
| Agentforce agents built in your org | Billed under existing Agentforce pricing | How your account team treats any external Headless Toolkit access |
| Custom scripts with fixed logic | Traditional integration | Whether the script wraps an LLM that picks the next call |
| HubSpot–Salesforce sync | Traditional integration | Whether any AI agent writes to Salesforce through a separate connection |
Two rows deserve extra attention. Agentforce already has its own metered usage types, such as actions and prompts, and the rate card notes that the new Headless Platform Interaction type applies to Headless 360 usage "in addition to" other usage types. And middleware can hide an agent. If Claude calls a MuleSoft API that writes to Salesforce, the call into Salesforce comes from the middleware, but the decision came from an agent. Salesforce hasn't said how it will treat that pattern, so flag it rather than guess.
Salesforce's registration model assumes each agent has its own identity. Many orgs today run agents under a shared integration user or a person's login. Fix that before registration opens.
The goal is accurate classification, not avoidance. Running agent traffic under an integration user to stay off the meter creates contract risk and leaves you without an audit trail showing which agent did what. For how external access and OAuth fit together, see our guide to Headless 360 architecture readiness.
Social posts have summarized the change as "for partners, this shows up at renewal." Salesforce's Help article is more specific, and the timing depends on how the agent connects and when you bought:
| Situation | What Salesforce says |
|---|---|
| Any use of a Salesforce MCP server | Register agents within three months of Salesforce's notice that Agentic Identity is available (all customers) |
| Existing customers using agents with 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 |
| Traditional integrations | Pricing and security options unchanged |
The renewal trigger in Salesforce's article applies to customers' API-based agents. It doesn't describe separate terms for ISV or AgentExchange partners, and MCP use isn't tied to renewal at all. Registration is done by the customer's admins in the customer's org, so a partner-built agent will likely follow the customer's timeline. Confirm that with your Salesforce account team and with each vendor. Salesforce Ben reports that Agentic Identity is expected to be generally available in November, in line with the Help article's November target.
Packaged apps add a wrinkle. Salesforce notes that for apps installed from an ISV's package, developer settings are locked, and enabling new features or changing scopes requires the vendor to release a new package version. If a vendor's feature needs to present registered-agent credentials, the vendor has to support it. Ask each vendor:
Ask your Salesforce account team the same classification questions for gray-zone patterns and get the answers in writing. For renewal-stage licensing questions, see our Agentforce AELA pricing decision guide. None of this is legal or contract advice. Review your agreements with your own counsel.
Add these fields to your integration inventory and fill them in for every production connection:
| Field | What to record |
|---|---|
| Connection name and owner | Business owner and technical owner |
| Connected app or External Client App | Name, OAuth flow and whether a managed package owns it |
| Runs as | Integration user, named user or registered agent |
| Who decides calls | Fixed logic, AI model or mixed |
| Protocol | Salesforce MCP server, REST, Bulk, Streaming or other APIs |
| Classification | Traditional, agentic or gray zone, plus the rationale |
| Registration deadline | MCP three-month rule, renewal date or not applicable |
| Call baseline | Calls per workflow or outcome over representative weeks |
| Permissions | Permission set group and objects touched |
| Vendor status | Answers to the renewal questions above |
Connected app usage, login history and Event Monitoring (where licensed) help you fill in "runs as" and call baselines.
Vantage Point runs integration reviews that sort traffic into agentic, traditional and gray-zone connections, then redesigns identities and permissions to match. Our MuleSoft integration team reviews middleware patterns where agents hide behind APIs, and our system integration and data migration practice covers ETL, sync and packaged-app connections. In one 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.
Not sure which of your connections Salesforce will treat as agents? Vantage Point can inventory your production connections, classify each one, and give you a list of identity fixes and vendor questions before your renewal. Schedule an agentic traffic review with Vantage Point.
Traffic is agentic when an AI agent decides at runtime which Salesforce calls to make, over MCP or a direct API. Salesforce will classify a connection as an agent when it presents OAuth credentials tied to a registered Agentic Identity. Traffic from designed integrations with a fixed sequence is treated as traditional integration traffic.
Scheduled or event-driven middleware flows with fixed logic are traditional integrations, and Salesforce says their pricing stays the same. The gray zone is middleware that exposes Salesforce as tools an AI agent calls. Document those patterns and ask your Salesforce account team to confirm the classification in writing.
Yes. An AI assistant that uses a Salesforce MCP server decides which calls to make, so it's agentic. Salesforce says any use of a Salesforce MCP server requires agent registration within three months of its Agentic Identity notice, for all customers.
Nothing in Salesforce's published guidance puts a standard field-mapping sync into the agentic category. It runs a designed sequence, which Salesforce describes as a traditional integration. Check separately for any AI agent that writes to Salesforce through its own connection.
It depends on how they connect. Agents using a Salesforce MCP server must be registered within three months of Salesforce's Agentic Identity notice. Existing customers' API-based agents must be registered and migrated to the new billing model at renewal. Customers purchasing on or after September 17, 2026, have three months from the notice for API-based agents.
Don't plan on it. Salesforce requires customers to register agents, and misclassifying agent traffic creates contract risk and weakens your audit trail. Give each agent its own registered identity with narrow permissions, and keep integration users for true system-to-system connections.
Ask whether any feature uses AI to choose Salesforce calls, whether it uses MCP or direct APIs, whose identity it runs under, and when it will support registered-agent credentials. For apps installed from a vendor's package, the vendor controls those settings, so support has to come from the vendor.
No. As of September 28, 2026, the Flex Credits Rate Card lists the Headless Platform Interaction multiplier as "TBA," and agentic usage isn't metered. Salesforce says it will give 30 days' notice before a multiplier is added and metering begins.
Vantage Point is a boutique CRM consulting firm helping businesses transform with Salesforce, HubSpot, and AI. Learn more at vantagepoint.io.