What is an MCP server?

The Model Context Protocol is an open standard that lets AI assistants such as ChatGPT and Claude call tools in your product. An MCP server is the part of your product that offers those tools, checks who is calling and decides what each caller may do.

Should the tenant ID be a tool parameter?

No. The server works out the caller's organisation from their verified token before any query runs, so an agent can never reach another customer's data by changing an argument.

Can the agent send emails on its own?

In AgentlyLeads, no. The agent can only create drafts. A person approves, and a separate send call checks the suppression list and the daily cap at the moment of sending, then records what was sent.

MCP Server Development for SaaS Products and CRMs

MCP server development

I build custom MCP servers that let ChatGPT, Claude and your own agents work inside your product. The server checks who is calling, works out their organisation before any query runs, and gives the agent tools that prepare work rather than act on their own. A person approves anything that reaches a customer.

Send a project brief Book a 20-min call

What goes into a production MCP server

Access checks on every call. The caller's token is verified first, and their organisation is resolved before any query, so data stays inside the right customer's account.
Tools shaped for safety. Agents get tools that read, search and draft. Anything that cannot be undone, such as sending a message, sits behind a separate step that a person approves.
Limits at the moment of action. Suppression lists, daily caps and an audit record are checked and written when the action happens, not when the agent first asks.
A path to production. The server ships with the rest of your product: deployed, monitored and documented.

Proof

I worked on AgentlyLeads, a CRM you run from ChatGPT and Claude, as an AI engineer on the CS2 Technologies team: the AI connection, the AI features, the security that keeps each customer's data separate, and the path to production. Its send flow is the pattern I bring to MCP work: the agent prepares, a person approves, and the system checks the rules again at send time.

This site runs its own read-only MCP server too, at agenticabdullah.com/mcp. Add it to Claude as a remote MCP server and ask about my work.

I hold the Anthropic certificates Model Context Protocol: Advanced Topics and Introduction to Model Context Protocol, and the PIAIC certificate Model Context Protocol Level 2 Developer.

When MCP is not the answer

MCP fits when an AI assistant should work inside your product. When your own AI product needs to reach many other systems, a shared connector can be the better design, as in one connector for HubSpot, Zoho and Jobber. Both sit under AI integration for existing systems.

Related services

/integrate

AI integration

Add AI sales agents, assistants, CRM integrations and human-in-the-loop workflows to live platforms, without rewriting what works or risking production data.

/ship

AI production deployment

Cloud infrastructure on AWS and Azure written as code, containers on Docker and Kubernetes, automated deployments with no stored passwords, secure secrets handling, load testing and clear handover documentation. I don't hand over a demo; I hand over a running system and the guides to operate it.