I will build a custom mcp server for claude, cursor or your ai agent


Over deze dienst
Your team wants an AI agent that works with your own systems: it reads your data, calls your API and acts within clear limits. Most demos stop before that point.
What you get:
- An MCP server in TypeScript that turns your API or data into tools an AI agent can call (Claude, Cursor, Codex or any MCP client)
- Read-only tools by default; tools that change data can wait for a human approval
- Setup docs so your team can run and extend it
- Premium: an agent with RAG search over your own documents
How it works:
1. You share the API docs or data source and the tasks the agent should do.
2. I send a short written plan with a fixed scope before any code.
3. Daily progress notes while I build, handover docs at the end.
Why me: 15+ years building software that runs in production. I built a self-hosted workspace that runs Claude Code, Codex, Gemini and OpenCode side by side, with MCP tool servers, browser automation (Playwright/CDP) and approval-gated autonomy. Stack: TypeScript, NodeJS, React, Fastify, PostgreSQL, Docker.
What I need from you: API docs or a schema, a test or sandbox account, and 3-5 example requests the agent should handle.
Maak kennis met Nikolaos
MCP and AI agent developer, Claude Code setup, Shopify ERP sync
- Afkomstig uitThailand
- Lid sindsokt 2026
Talen
Grieks, Engels
Mijn portfolio
Andere AI-development diensten die ik aanbied
Veelgestelde vragen
What is an MCP server?
MCP (Model Context Protocol) is an open standard that lets AI agents call tools. An MCP server wraps your API or data as tools with clear inputs, so Claude, Codex, Gemini and other MCP clients can use them.
Which stack do you use?
TypeScript on NodeJS, with PostgreSQL when the server needs storage. If your team works in another stack, message me before ordering.
Can the agent change data in my system?
Only if you want it to. Tools are read-only by default. Tools that write, delete or send can require a human approval before they run.
My project is bigger than Premium. What then?
Send me the scope. I will propose fixed-price milestones with a written plan. Work on each milestone starts when it is funded.
How do we communicate?
Through Fiverr messages. I am in Bangkok (UTC+7) and overlap with EU mornings and US evenings. You get a short progress note every working day.
What counts as one tool?
One action the AI can take, such as "search orders" or "create a draft invoice", with typed inputs and clear errors. A package's tool count is the number of these actions.
Which AI clients will it work with?
Any MCP client: Claude Desktop, Claude Code, Cursor, VS Code, Codex and others. I include the config for the clients you use.
Does it run locally or on a server?
Basic runs locally over stdio. Standard and Premium add a remote HTTP endpoint with auth, on your server or in Docker.
Do I own the code?
Yes. You get the full source in your repository, with a README and a setup guide. No license fees and no lock-in.
How do you test it?
Every package is checked with your example requests before delivery. Standard and Premium also include automated tests for each tool.

