Browse categories
Explore
Fiverr Pro
English
$
USD
I will build a custom mcp server for claude and your ai agents
Afriza Gilleon
About this gig
Your API already works. The problem is that an AI agent cannot use it safely. It guesses parameters, calls the wrong endpoint, or invents data when a response comes back empty.
What I build for you:
- A typed MCP server over your existing API, database, or internal service
- Read-only by default, with any write operation guarded so two agents cannot overwrite each other
- Errors that return a usable message instead of silence
- Tested against a real client, not just unit tests
- A short setup document your developer can follow
Why me: I built a self-hosted workflow platform that exposes 13 MCP tools to AI agents, and a benchmark that measured how often AI output only looks correct. It found that 7% of one agent's accepted results were fake: they matched recorded data and broke the moment inputs changed. That failure mode is what I design against.
Stack: TypeScript and Node.js, or Python if you prefer. I work UTC+7 and reply the same day.
Not sure whether your API fits? Message me with what it does and I will tell you honestly before you order.
Get to know Afriza Gilleon
Afriza Gilleon
MCP servers and automation that AI agents can actually run
- FromIndonesia
- Member sinceAug 2026
- Avg. response time1 hour
Languages
English
I build the tooling AI agents run on. I built a self-hosted workflow platform that exposes 13 MCP tools so agents can read and run workflows safely, a browser automation runtime whose checkpoints survive page reloads, and a benchmark that found 7% of an AI agent's accepted code repairs were fake — they matched recorded output and broke on new inputs. TypeScript, Node, React, PostgreSQL, Python. I tell you what a tool cannot do as clearly as what it can, so you can plan around real limits. Same-day replies, UTC+7.
FAQ
Do you need access to our production system?
No. I work against your API documentation and a test or staging endpoint. If credentials are needed, use a read-only key scoped to test data. I never ask for production credentials.
Which MCP clients does this work with?
Anything that speaks MCP — Claude Desktop and Claude Code, and other MCP-compatible clients. I test against a real client before delivery, not just unit tests.
What if our API changes later?
Typed tool definitions make that a small edit rather than a rebuild. I can quote a maintenance update, or hand over documentation clear enough for your own developer to do it.
