WHMCS MCP servers: what they do and how to pick one safely
An MCP server lets an AI assistant such as Claude or ChatGPT read and act in your WHMCS. Most are built for admins and one for clients. Here is how they differ, and the safety features to check before you install one.
MagizAI Team6 min read
A WHMCS MCP server lets an AI assistant such as Claude, ChatGPT or Cursor use WHMCS as a set of tools, to find a client or draft a ticket reply. Most are for admins; one we found is for clients. They differ in where they run, how they sign in, what each token may do and what needs approval.
What is MCP?
MCP stands for Model Context Protocol. Its own site calls it "an open-source standard for connecting AI applications to external systems" and compares it to "a USB-C port for AI applications" (modelcontextprotocol.io, read on 19 October 2026). A server publishes a list of tools, each with a name and the arguments it takes. The assistant decides when a question needs one, calls it and reads the result.
For WHMCS, that means someone can ask Claude which clients have invoices more than 30 days overdue and get an answer built from your real data, with no report to write. It also means the assistant can act on that data, and that is where the risk sits.
Who is the AI working for: you or your client?
Almost every WHMCS MCP server is an admin tool. The assistant usually works with staff-level reach across all your clients, so one bad call can touch any account.
Provonic's Clients MCP Server is the exception we found. It lets a client connect their own AI assistant to their own account, and its Marketplace listing says "Every action runs as that client, on that client's own records, with WHMCS user permissions still applied." It lists 179 tools, 96 of them for cPanel, and says "Anything that costs money or cannot be undone needs the client's approval in the chat first." The risk changes shape: one account at a time, but driven by people you never trained.
WHMCS MCP servers on the Marketplace compared
Each row comes from the product's WHMCS Marketplace listing, checked on 19 October 2026. Nothing was installed or tested, and "not stated" means the listing is silent on it.
| Server | Price on the listing | Tools | Sign-in and limits | Needs a person's yes | Version |
|---|---|---|---|---|---|
| Provonic WHMCS MCP Server | $7.50 a month, $39 a year, $75 one-time, $129 with source; 5-day trial | "140+ native tools" | Self-hosted; wmcp_ tokens stored as SHA-256 hashes, per-token permissions, rate limits, audit log | Not stated | 1.0.0, 30 Jun 2026 |
| Provonic Clients MCP Server | $7.50 a month, $39 a year, $59 one-time, $129 with source | 179 | Runs as the client; an API key or optional OAuth | Anything that costs money or cannot be undone, approved by the client | 1.0.0, 14 Oct 2026 |
| MX Modules MCP Server | $22.00; the vendor's site says $22 a month | 68 in the core, 84 with two companion addons | API keys with read or write per key, rate limits, audit log; works with Ollama | Not stated | 1.7.1, 13 May 2026 |
| PD Modules MCP Hub | Free, "no paid tier" | Not counted | A token per connection, each with a ceiling from read only to dangerous; acts as an admin you choose; kill switch | Refunds, payments, cancellations, ticket replies and anything that emails a customer | 0.1.26, 23 Sep 2026 |
| Relyagent | $59.00, term not stated | Scope-gated sets, switchable per connection | OAuth 2.0 with PKCE, hourly and daily quotas per connection, tamper-evident audit log | Not stated | 1.0.0, 25 Aug 2026 |
| Daddario WHMCS MCP Server | $14.99 a month or $129.99 a year; 14-day trial | 96 on the listing | A Docker container outside WHMCS; needs a WHMCS API credential | Not stated; offers a dry run on every write | 1.0.0, 5 Sep 2026 |
Daddario's own README on GitHub gives two other tool counts, 98 and 86, so ask the vendor which one you are buying.
Are there free, open-source WHMCS MCP servers?
Yes, on GitHub and npm. caioldcarvalho/whmcs-mcp (MIT licence) says it has "34 MCP tools implemented and tested" and signs in with a WHMCS API identifier and secret, or an admin username and password. The npm package whmcs-mcp-server, version 1.3.1 of April 2026, describes "60+ tools" while its README totals 73. Both talk to the WHMCS API from outside, so both need an API credential, and you are the one who hosts, patches and watches them.
What safety features should a WHMCS MCP server have?
- A token per assistant, with its own permissions. Provonic, MX Modules and PD Modules all list this. One shared key for everything is the setup to avoid.
- A person's yes before money moves. PD Modules holds refunds, payments, cancellations, ticket replies and "anything that emails a customer" for approval. Provonic's client edition asks the client.
- A dry run. Daddario says every write can run in dry-run mode first, so you see what would change before it does.
- An audit log you will actually read. Most list one. Check that it records the tool, the arguments and the result, not only the time.
- A kill switch. PD Modules cuts every connection from its panel, or when a file called .mcp-kill is dropped into its folder.
- Local models when data must stay put. MX Modules says it works with Ollama, so prompts and WHMCS data can stay on your own machines.
- No client-area exposure for admin tools. PD Modules puts it plainly: "Client Area Features None, permanently."
Be careful with the WHMCS API Access Key. Daddario's README recommends one because "An Access Key bypasses IP restrictions entirely", and that is exactly why it widens the door. If a server needs an API credential, tie it to an API role that allows only the actions it uses.
Is an MCP server the same as a chatbot for your clients?
No. An admin MCP server answers to a member of staff who can already see everything. A chat in your client area answers strangers and clients, so it has to prove who is signed in, look up only that client's records and confirm each change with them. Those are different safety problems, and a tool built for one does not solve the other.
MagizAI works on the client side, and it is not a WHMCS MCP server: its WHMCS AI chatbot module reaches WHMCS through a signed bridge rather than MCP. It is an MCP client, though, so you can connect your own MCP server for other systems and tick the tools the assistant may call. The MCP guide lists the limits: a public https address using Streamable HTTP, no local servers, and no servers that sign in only with OAuth yet, which rules out connecting a server like Relyagent today.
How to choose a WHMCS MCP server
- Decide who will use it: staff working in Claude or ChatGPT, or clients with their own assistant.
- Start with a read-only token and a week of real questions before you allow any write.
- Turn on writes one area at a time, with approval for anything that costs money or emails a client.
- Read the audit log every week, and make sure everyone on the team knows where the kill switch is.
- Check the listing's version history against your WHMCS version before you buy.
For the other ways to bring a model into WHMCS, see how to add ChatGPT, Claude or Gemini to WHMCS. Before any tool gets write access, run through what to check before you let AI touch a customer account. And if you want to connect an MCP server to a chat on your website, connecting tools with MCP covers it step by step.