I will build custom mcp servers for claude, cursor and your ai agents


About this gig
Your AI can only act on what it can reach. An MCP server is the bridge: it gives Claude, Cursor, ChatGPT or your own agent safe, structured access to your database, your API, your files and your internal tools.
WHAT I BUILD
- Custom MCP servers in Python or TypeScript for Claude Desktop, Claude Code, Cursor and any MCP client
- Tools that read and write your real systems: databases, REST APIs, CRMs, spreadsheets, documents
- Authentication, permission scoping and audit logging, so the model can only do what you allow
- Remote hosting on your infrastructure, with monitoring
WHY ME
I run an AI company and build agents that operate in production. MCP is how I connect them to real systems, and I use it daily in my own work. You get a server that survives real data, not a demo.
HOW IT WORKS
1. Tell me which system the AI should reach and what it may do
2. I design the tools and confirm scope
3. I build, test against your real environment, and hand over
4. You get the source, a setup guide and a walkthrough
Every tool has a dry-run mode and explicit error handling. Nothing acts silently.
Message me with the system you want your AI to reach and I will tell you what it takes.
Get to know Thejan
AI Engineer Chatbots, AI Agents and Workflow Automation
- FromSri Lanka
- Member sinceJan 2024
- Last delivery4 months
Languages
English, Sinhala
My Portfolio
Other AI Development Services I Offer
FAQ
Which AI tools can use the MCP server?
Any MCP-compatible client: Claude Desktop, Claude Code, Cursor, Windsurf, ChatGPT with MCP support, and custom agents built on the OpenAI or Anthropic SDKs. I test in the client you actually use before handover.
Do I have to give you my passwords or API keys?
No. You create scoped credentials yourself and grant only what the tools need, revocable at any time. I build against a sandbox first and only connect production once you have seen it work.
Can the AI change or delete data in my systems?
Only if you want it to. Read-only is the default. Every write tool is explicit and separately permissioned, and each ships with a dry-run mode that logs what it would do without doing it, so you can watch it before you let it act.
Where does the MCP server run?
Locally on your machine for Claude Desktop, Claude Code or Cursor, or hosted remotely on your own cloud for teams and agents. Standard includes deployment to your host. Premium includes monitoring.
Python or TypeScript?
Whichever fits your stack, I build in both. Python if your systems already live in Python or you need data tooling; TypeScript if you want the tightest fit with the official SDK and the editor clients. I will recommend one after seeing your environment.
Do I own the code?
Completely. You get the full source, a setup guide and a walkthrough video, running on your own accounts and hosting. No licence, no lock-in and no ongoing payment to me.
What happens when an API changes or a tool breaks?
Every tool has explicit error handling and logging, so failures are visible rather than silent. Premium includes 30 days of support after delivery, and I offer a monthly maintenance option after that if you want one.
Can you connect a system that has no API?
Often, yes. Databases, files, spreadsheets and internal tools can be reached directly. For a web app with no API I can build a controlled browser-driven tool. Send me the system before you order and I will confirm what is possible.

