Plug WHMCS into any AI agent.
Relyagent turns your WHMCS install into a self-hosted MCP server. Claude Desktop, Cursor, your custom bot — they speak Model Context Protocol; now WHMCS does too. OAuth-gated, scope-controlled, audit-logged.
Clients · Tickets · Invoices
Bearer-token auth
Read & write split
With param redaction
Per-token windows
One bridge. Every agent, every tool.
AI assistants connect through the Relyagent MCP gateway — OAuth-secured, scope-gated, rate-limited and fully audited — to reach your WHMCS clients, tickets, invoices, services and domains.
Simple, transparent licensing
One license per WHMCS installation. Includes updates, support, and unlimited MCP clients & tool calls.
Leased License
Yearly plan with continuous updates
- One WHMCS installation
- Unlimited MCP clients & tool calls
- One-click auto-updates
- Email & ticket support
Lifetime License
Pay once, own it forever
- One WHMCS installation
- Unlimited MCP clients & tool calls
- 1 year free support & updates
- Email & ticket support
Everything an LLM needs. Nothing it shouldn't.
A complete MCP server built into your WHMCS — with the guardrails your business actually needs in production.
6 tool groups, 25+ tools
Clients, tickets, invoices, services, domains, admin. Read or write, all scope-gated. Toggle entire groups on or off from the admin panel.
OAuth 2.0 + PKCE server
Full OAuth authorization-code server built in. Register clients from the admin UI, consent-screen flow, RSA-signed JWTs. No third-party auth service needed.
Streamable HTTP + SSE
Both MCP transports out of the box. Auto-negotiated per request via the Accept header. Works with every spec-compliant client.
Prompt-injection sanitiser
Ten known jailbreak patterns scrubbed from ticket bodies, client notes, and invoice descriptions before they reach the model. Detected hits flagged in the audit log.
Per-token rate limiting
Fixed-window minute + day counters per access token. Defaults 60/min and 5,000/day, tunable per install. Returns HTTP 429 + Retry-After on overflow.
Audit log + dangerous-tool gate
One row per tool call — success, sanitized, forbidden, rate-limited, error. Sensitive params redacted. Write tools refuse to fire until you flip the admin toggle.
Granular consent. No magic. No surprises.
Every MCP client is registered explicitly. Every connection goes through an admin consent screen with the exact scopes laid out. PKCE-protected, one-time auth codes, refresh tokens with rotation tracking.
- 11 scopes —
clients:read,tickets:write,admin:read, etc. - Per-client scope allow-list, enforced at token mint
- Refresh-token rotation with revocation tracking
- One-click panic switch revokes every active token
clients:readView client profilestickets:readView support ticketstickets:writeReply, change statusadmin:readStats & activity logSee every connection. Revoke any of them.
The Relyagent admin panel gives you a live view of every registered client, every active token, and every tool call. Three KPIs at a glance, a click-to-revoke table, and a one-button panic switch when something feels off.
- Live counts — connected clients, active tokens, calls today
- One-time secret reveal on client registration
- Per-client revoke + global "revoke all tokens" panic switch
- Group toggles for clients / tickets / invoices / services / domains / admin
Every tool call. Every parameter. Every status.
Relyagent writes one audit row per tool call — regardless of outcome. Success, sanitised, forbidden, rate-limited, or error. Sensitive parameters (passwords, card numbers, OAuth secrets) are redacted before insert. Queryable from the admin panel, exportable as CSV.
- One row per dispatch — success / sanitized / forbidden / rate_limited / error
- 20+ redacted param keys — passwords, tokens, card data
- Filterable by tool name, status, and date range
- CSV export with no-cache headers — up to 10k rows
Built assuming an LLM will misbehave
AI agents make mistakes. Sometimes they get prompt-injected. Sometimes they hallucinate a destructive action. Relyagent assumes the worst — and contains it at four layers.
OAuth 2.0 + PKCE
Bearer-token auth via league/oauth2-server. RSA-signed JWTs. Per-install keypair, rotatable on demand. Every call validated through a real resource server.
Every dispatch logged
One audit row per tool call. Success, sanitised, forbidden, rate-limited, or error — all captured. Sensitive parameters redacted before insert. CSV export, no leaks.
Per-token rate limiting
Minute and day windows per access token. Defaults 60/min and 5,000/day. Tunable up to 10,000/min, 1M/day. HTTP 429 + Retry-After on overflow.
Prompt-injection defence
Ten known jailbreak patterns scrubbed from every value flowing back to the LLM. Hits flagged in the audit log as status=sanitized for forensic review.
Requirements & compatibility
Relyagent runs on standard WHMCS environments. No exotic dependencies, no extra services to manage, no third-party auth provider.
System Requirements
PHP Extensions
What is Relyagent in one sentence?
A WHMCS addon that exposes your WHMCS install as a self-hosted MCP (Model Context Protocol) server — so any LLM client that speaks MCP (Claude Desktop, Cursor, your custom agent) can call WHMCS tools through OAuth 2.0 + PKCE.
Roughly 25 scope-gated tools across six groups: clients, tickets, invoices, services, domains, and admin.
How does an LLM client connect?
You register the client from Addons → Relyagent → MCP Clients, picking which OAuth scopes it can request. The client then uses the standard OAuth 2.0 authorization-code flow with PKCE — you approve the request on a consent screen with the exact scopes spelled out, and the client receives a bearer access token and refresh token.
Every subsequent JSON-RPC call carries that bearer. Refresh tokens rotate on use.
Does it work with Claude Desktop, Cursor, and the Anthropic API?
Yes — Relyagent ships both MCP transports: Streamable HTTP (default) and Server-Sent Events (when the client sends Accept: text/event-stream). Anything that implements the MCP 2025-03-26 spec works out of the box.
Is it safe to expose WHMCS data to an AI agent?
That's the entire design centre. Every tool requires a specific OAuth scope. Write operations require an additional "Dangerous Operations" admin toggle, which starts off. Every parameter and result flows through a prompt-injection sanitiser. Every call is rate-limited per token. Every call is audit-logged with sensitive parameters redacted.
The agent can only do what you explicitly granted — and you can see exactly what it did.
Can I disable write operations entirely?
Yes — and that's the default. The Dangerous Operations toggle in MCP Settings starts off. Until you flip it on, write tools (suspend_service, create_invoice, update_client, add_credit, etc.) reject every call with MCP error -32004.
You can also disable entire tool groups (clients, tickets, invoices, services, domains, admin) independently — useful for staging environments or restricted-purpose agents.
What about rate limits and abuse prevention?
Each access token has its own minute + day counters. Defaults are 60/min and 5,000/day, tunable in MCP Settings (hard ceilings: 10,000/min and 1,000,000/day). On overflow the request returns HTTP 429 with a Retry-After header and MCP error -32029.
There's also a one-click "Revoke All Tokens" panic switch in the admin panel for when something's clearly wrong.
How does the prompt-injection sanitiser work?
A library of ten known jailbreak patterns — ignore previous instructions, you are now a/an, act as, [INST], <|im_start|>, DAN mode, jailbreak, and several others — is matched against every string in every tool response. Anywhere user-generated text lives: ticket bodies, client notes, invoice descriptions, etc.
Matches are replaced with [sanitized]. The audit row gets tagged status=sanitized so you can spot which calls hit a pattern.
What WHMCS versions are supported?
Relyagent v1.0.x is built and tested for WHMCS 8.x with PHP 7.4 or higher. ionCube Loader is required (it's pre-installed on virtually all WHMCS-ready hosting environments, including all Relyweb hosting plans).
How do licensing and updates work?
One license per WHMCS installation, matched by domain and directory. The plugin checks your license periodically and caches the result locally. Updates are checked daily — when one's available, click "Update Now" from the admin panel; the plugin backs up the current installation before applying the update and auto-restores if anything fails.
Annual licenses continue to work after expiry, but stop receiving updates and support. Lifetime licenses never expire.
Can I get a refund if it doesn't work for me?
Yes — we offer a 14-day money-back guarantee. If Relyagent doesn't fit your setup or doesn't meet your needs, just contact support within 14 days of purchase for a full refund.
Release history
What’s new in Relyagent — newest first.
v1.0.0 08 Sep 2026 Latest
Improvements
- The module now includes a more reliable system for checking and installing future updates.
Fixed
- The system is now more stable when it cannot reach or read the server.
- Security has been improved for the offline feature gate and live update checks.
Related apps
Other Relyweb apps that fit alongside Relyagent.
Flint
AI chatbox, universal text rewrite, KB-grounded coach personas.
GST Manager
GST compliance for WHMCS — GSTR-1, export invoices, audit trails.
RelyPay
One payment layer for every gateway — Razorpay, Cashfree, PhonePe, more.
Hand WHMCS to your AI agents. Safely.
OAuth-protected. Audit-logged. Scope-controlled. 14-day money-back guarantee, no risk.