MCP server development
The Model Context Protocol (MCP) lets AI assistants use your software safely. I build MCP servers that expose your CRM, database, internal tools or even a browser as tools an agent can call, with least-privilege access and approval gates.
What an MCP server is, in plain words
An MCP server is a small program that tells an AI assistant what it can do in your systems: look up a customer, create a quote, run this report. The assistant (Claude, ChatGPT or your own agent) calls those tools. You decide exactly which tools exist and what each one may touch.
When you need one
- You want Claude or ChatGPT to work with your CRM, database or internal tools
- You are building your own agent and want a standard way to give it tools
- You want an agent to operate a web app that has no API, using browser tools
- You want one integration that works across several AI apps
What I build
- Read-only tools first (look-ups and reports), then write tools with approval
- Authentication and per-user permissions
- Rate limits, logging and an audit trail
- Clear tool descriptions so models choose the right tool
- Local and remote (hosted) servers
- Documentation and examples for your team
Proof: MCP in my own products
PagePilot gives Claude browser tools over MCP: navigate, read, click, type, upload and screenshot, with approval gates, site restrictions, a pause switch and action logs. MyTest uses browser tools to run AI testing agents. ros2-isaac-agent-suite is an open-source Claude Code plugin for testing robot software in simulation.
Security basics
- Least privilege: each tool can do one thing and nothing else
- No secrets inside prompts
- Approval steps for sending, paying and deleting
- Logs for every call, and a staging setup for testing
Frequently asked questions
What is the difference between an MCP server and an API?
An API is built for programmers and documented for humans. An MCP server describes its tools in a form an AI model can read, so the model can decide which tool to call, within limits you set. Many MCP servers sit on top of an existing API.
Which AI apps work with MCP?
MCP is an open standard supported by a growing list of AI apps and agent frameworks, including Claude. A server built once can be used by any compatible app.
Is it safe to let an AI use my systems through MCP?
It is as safe as you design it. I use least-privilege tools, approval steps for risky actions, audit logs and separate staging, so the AI can only do what you explicitly allow.
Can an MCP server connect to my CRM or WhatsApp?
Yes. If your CRM, database or messaging provider has an API, an MCP server can expose selected actions from it as tools.
How long does an MCP server take to build?
A simple server with a few read-only tools can take days to a couple of weeks. More tools, permissions and write actions take longer. I give an estimate after scoping.
Keep reading
Tell me what eats your week.
My intake agent asks a few quick questions and puts your brief in front of me. I read every one.