Claude, ChatGPT or an IDE, connected to your org
Claude
ChatGPT
IDEs and agents
MCP gateway
identity, scope, audit
Your Salesforce org
Records, metadata and deploys, under the permissions you already run.
This article gives you the ranges. We will map your number.
What MCP costs in your org depends on the workflow. Send it and GAT maps which calls stay inside the API capacity you already own, which become metered actions, and what the monthly credits math looks like, with the assumptions you should correct marked.
Prefer to see the mechanics first? Model the credits in the calculator or see the systems review.
Get an MCP cost map for your workflow
Send the workflow and expected volume. We reply with the free-versus-metered split and the first questions we would answer.
By submitting, you agree that GAT Solutions may contact you about this request. We do not sell your information.
Price the whole setup: who will use it, what Salesforce information it can access, what it can change, who reviews important actions, what happens when something fails, and who supports it after launch.
GAT Solutions sets up, builds, hosts, and maintains MCP servers (Salesforce Hosted MCP for your org, and custom MCP servers for the systems around it) connected to Claude, ChatGPT, or whichever assistant your team uses. We also run our own MCP product in production, so the guidance below comes from operating this category, not summarizing it.
Where the money goes
Hosted MCP access
Included
Comes with Enterprise Edition and above. For many orgs there is no MCP SKU to buy at all.
Standard record operations
Existing API limits
Reads, queries, and existing Apex and Flows draw on capacity your plan already includes.
Metered features, when the workflow uses them
$0.10 per standard action
Agentforce, Data 360 and Prompt Builder still bill usage. $500 buys 100,000 Flex Credits at list, and credits expire at the Order End Date.
Setup, permissions, review, testing, enablement
The real budget
Who connects, what they can see and change, which writes need approval, and how the team learns the workflow. Scoped per org, which is why we will not publish a band for it.
What Salesforce Hosted MCP costs
Salesforce's hosted MCP servers connect compatible AI clients (Claude, ChatGPT, IDEs) to your org under Salesforce's management. There is no single price because different parts of the workflow are billed differently:
- Access is included with Enterprise Edition and above. Salesforce's documentation states Hosted MCP access comes with Enterprise Edition and higher. For many orgs there is no separate MCP SKU to buy. The first budgeting step is simply checking what your current Salesforce plan already includes before buying anything new.
- Normal Salesforce actions use your existing API limits. Reading records, running queries, and using existing Apex and Flows draw on the API capacity your Salesforce plan already includes. They do not create a new charge by themselves. Salesforce's standard hosted servers focus on exactly those operations, and its documentation currently lists the new Hosted MCP usage types (Record Operation, Process Invocation) as TBA and not yet metered.
- Usage-based Salesforce features still charge for usage. If the workflow uses Agentforce, Data 360, Prompt Builder, or another consumption-based product, that product still charges the normal usage fee. This is where Flex Credits enter the budget.
- "Included" is not "set up." The endpoint being available is the beginning. Who connects, what they can see and do, how writes are controlled, and how people learn the workflow still need to be decided, configured, tested, and documented. That hands-on work is often the larger part of the project.
Flex Credits, in dollars
Flex Credits are Salesforce's prepaid usage currency for metered products. The published list price is $500 per 100,000 credits ($0.005 each), and every metered operation consumes a set multiplier from the published rate card: a standard Agentforce action uses 20 credits ($0.10 at list) in production and 16 in a sandbox, a voice action 30, and Data 360 operations draw per million rows or per megabyte, a spread that runs from 3 credits to 75,000 depending on the operation. Credits expire at the Order End Date with no rollover.
The budgeting trap is that one customer or employee request can trigger several paid actions. A conversation may run more than one prompt, update records, call another Salesforce feature, or use voice, so the total can be higher than the price of one action. Before approving the project, estimate how often people will use the workflow, which paid features each request will call, and what that means in dollars. Confirm the final terms with your Salesforce account team because packaging changes frequently.
Teams usually bring us in to answer three practical questions:
- What does our current Salesforce plan already include? We review the relevant plan and order details before the team buys another product.
- What will normal usage cost? We map the steps behind a typical request, estimate how often they will run, and compare the expected usage with Salesforce's published prices.
- Which steps do not need a paid AI feature? Many useful jobs are ordinary Salesforce searches, record updates, Apex, or Flows that use the API limits you already pay for. Keeping simple work simple can lower the bill.
(Packaged MCP products also exist at flat per-user pricing, including ApexGenius, and can make sense when usage-based pricing does not fit. The main question is not which technical label sounds best. It is which option does the work your team needs at a cost you can predict.)
Or skip the build
Use ApexGenius, the Salesforce admin MCP
We built the connector this article prices out. Flat monthly rate, no per-action fees, 7-day free trial: Claude and ChatGPT administer and develop your org with staging, logging, and human-approved deploys.
What are you paying a consultant to do?
A useful proposal should list the actual work. If it only has one line called "MCP setup," ask whether it includes each item below:
- A clear workflow: who will use it, what they need to do, which systems are involved, and what a good result looks like.
- Sign-in and account setup: whether each person uses their own Salesforce account or the workflow uses a shared account, plus how access is removed when needed.
- Salesforce user access: which records and fields each person can see, change, create, or delete.
- Allowed AI actions: what AI can read, update, build, test, or deploy, and what stays blocked.
- Human review: which changes need a preview, approval, or final sign-off before they happen.
- A plan for failures: what happens when access is denied, information is missing, Salesforce rejects a change, or another system is unavailable.
- Real testing: normal requests, wrong requests, blocked requests, and important changes tested before launch.
- Training and documentation: clear examples for each role, rules for safe use, and a named person to contact for help.
- Ongoing ownership: someone responsible for reviewing access, updating the workflow, and fixing problems after launch.
The six factors that move implementation cost
- Number of systems. Salesforce-only is different from a workflow spanning Salesforce, HubSpot, support, documents, billing, and internal APIs.
- What AI is allowed to do. Answering a question is simpler than changing customer records, making a case decision, triggering a financial action, or changing how Salesforce is built.
- How many people and access levels are involved. One Salesforce admin is simpler than several teams with different roles, company sign-on requirements, shared system accounts, or regulated data.
- Whether company-specific actions must be built. A ready-made connection is already software. A new action built for your business needs design, code, testing, alerts, documentation, and support.
- How much testing and review the launch requires. Security reviews, formal test plans, audit requirements, backup procedures, and launch approvals add work because important mistakes become less likely.
- How many people will use it. A pilot for one admin is much simpler than rolling Claude or ChatGPT out to an entire company, even if both use the same Salesforce connection.
The MCP work we do
Everything on the checklist above is work we sell, as a project, or as ongoing ownership. Four ways clients use us:
- Set up Salesforce Hosted MCP. We confirm what your edition already includes, connect Claude or ChatGPT, design who can see and change what, test the real workflows, and train the team. Usually a fixed project, delivered in weeks.
- Build custom MCP servers. When the work spans more than Salesforce (billing, an EHR, a booking platform, internal APIs), we build the MCP server that puts those systems safely in reach of an AI assistant: read-only by default, risky actions behind an approval, nothing exposed by accident. We have shipped this in production for a company that runs daily operations through it.
- Host and run them. A custom MCP server is software that has to live somewhere and stay up. We host and operate MCP servers for clients (monitoring, upgrades, and someone to call) so using one does not require standing up infrastructure or hiring for it.
- Maintain and govern, ongoing. Access reviews as people join and leave, new tools as workflows grow, updates when a vendor changes an API, and an eye on usage costs before they surprise anyone. This is the monthly-ownership version, and it is how most launches should end.
What kind of help do you actually need?
Most companies fall into one of three situations. The difference is what you want AI to do, how many systems are involved, and how much help your team needs to set it up and run it safely.
1. You can probably set it up yourself
You mainly want Claude or ChatGPT to work with Salesforce, your Salesforce plan or a ready-made connector already covers the task, and you have an admin who can manage user access. We can check what you already have and give you a setup checklist. If you do not need a consulting project, we will tell you.
2. You need help launching it safely
The connection works, but your team needs help deciding who can use it, which Salesforce information AI can see, what it is allowed to change, and how to test everything before it reaches your live org. We set up the connection, configures access, tests your real workflows, documents the setup, and trains your team. This is usually a fixed project delivered in weeks.
3. You need something built for your business
You want AI to work across Salesforce and other systems, carry out company-specific tasks, search internal documents, or ask a person for approval before taking an important action. That is a custom software project. We design and builds the complete workflow, including the safety checks, testing, documentation, and handoff. We have shipped this kind of system in production .
Need help deciding which situation you are in? See how we help teams connect Claude or ChatGPT to Salesforce, see the Claude Salesforce integration and ChatGPT Salesforce integration guides, or compare the connection options side by side.
Does Salesforce Hosted MCP mean the implementation is free?
No. Included product access and zero implementation work are not the same thing. Your Salesforce plan may include the connection, but your team may still need help connecting Claude or ChatGPT, setting user access, deciding what AI can change, testing the workflow, documenting it, training users, and supporting it after launch.
Should we start with Agentforce or MCP?
Start with where people will use the AI. If your team wants to work from Claude, ChatGPT, Cursor, or another outside AI tool, an MCP connection is usually the better place to start. If customers or employees should use an AI assistant inside Salesforce, on a website, or in a service channel, Agentforce may fit better (we break down the licensing math in Agentforce pricing, explained). Some companies use both for different people and tasks.
If the workflow invokes metered Agentforce actions, use the Salesforce Flex Credits rate card to see the current multipliers in dollars, then test the action volume in our free Flex Credit calculator.
Where should we go deeper?
If you are scoping this work, three companion guides carry the detail: how to connect Claude to Salesforce walks the setup end to end, and the day-two operating guides show what teams actually do with the connection: Salesforce admin tasks with Claude and Salesforce admin tasks with ChatGPT.
Can you give us an exact implementation price?
Yes, after the users, systems, actions, environments, controls, and launch scope are clear. We do not publish invented price bands that pretend a one-user setup and a regulated multi-system agent are the same service. Plan your Salesforce + AI rollout and we will recommend the smallest option that does the job.