You will get a custom MCP server in Python so AI agents can use your tools
Project details
An MCP server is the piece that lets AI agents like Claude or Cursor actually use your API, your
database or your internal tools - instead of only talking about them.
I build them in Python with FastMCP, with typed and validated inputs, so they hold up when an
agent sends something unexpected. That last part is where most MCP servers fall over, and it is
what I specialise in: I wrote the open-source tool that audits MCP servers for exactly this, and
I have run it against 13 public servers including ones from Anthropic, Google and Microsoft.
WHAT YOU GET
• A working MCP server exposing your functionality as typed tools
• Input validation that rejects bad arguments cleanly instead of crashing
• A test suite that runs without needing a live server
• Docker setup and documentation
• Full source code, yours to keep and modify
• An audit report from my own tool, showing it handles malformed input correctly
WHAT I NEED FROM YOU
Your API docs, or a description of what the agent should be able to do. Test credentials if it
talks to a real service. I will send you a tool list to approve before I start building.
github.com/junaidshahid-dev/mcp-probe
database or your internal tools - instead of only talking about them.
I build them in Python with FastMCP, with typed and validated inputs, so they hold up when an
agent sends something unexpected. That last part is where most MCP servers fall over, and it is
what I specialise in: I wrote the open-source tool that audits MCP servers for exactly this, and
I have run it against 13 public servers including ones from Anthropic, Google and Microsoft.
WHAT YOU GET
• A working MCP server exposing your functionality as typed tools
• Input validation that rejects bad arguments cleanly instead of crashing
• A test suite that runs without needing a live server
• Docker setup and documentation
• Full source code, yours to keep and modify
• An audit report from my own tool, showing it handles malformed input correctly
WHAT I NEED FROM YOU
Your API docs, or a description of what the agent should be able to do. Test credentials if it
talks to a real service. I will send you a tool list to approve before I start building.
github.com/junaidshahid-dev/mcp-probe
AI Algorithms
Regression AnalysisAI Applications
AI-Generated CodeAI Development Language
PythonAI Models
OpenAI CodexWhat's included
| Service Tiers |
Starter
$250
|
Standard
$600
|
Advanced
$1,200
|
|---|---|---|---|
| Delivery Time | 3 days | 5 days | 7 days |
Number of Revisions | 1 | 2 | 2 |
AI Model Integration | - | - | - |
Batch Normalization | - | - | - |
Database Integration | - | - | - |
Detailed Code Comments | - | - | - |
Image Upscaling | - | - | - |
MLOps | - | - | - |
Model Deployment | - | - | - |
Model Documentation | - | - | - |
Model Monitoring | - | - | - |
Model Testing & Optimization | - | - | - |
Model Tuning | - | - | - |
Natural Language Processing | - | - | - |
NLP Tokenization | - | - | - |
Pre-Training | - | - | - |
Prompt Engineering | - | - | - |
Setup File | - | - | - |
Source Code | - | - |
Optional add-ons
You can add these on the next page.
Fast Delivery
+$10 - $800About Junaid
MCP Servers & AI Agent Tools | Python, FastAPI | I also audit them
Lahore, Pakistan - 7:32 pm local time
Lately, most of my work has been with LLMs and AI agents:
- MCP servers: the piece that lets AI agents like Claude actually use your tools and data
- Auditing MCP servers: finding out what breaks before your users do
- RAG systems: "chat with your documents" assistants that answer from your files and cite the
exact source, so nothing gets made up
- Python / FastAPI / Flask backends and REST APIs
Here is the part that makes me different from everyone else on this page.
I wrote mcp-probe, an open-source tool that audits MCP servers. It connects to a server,
discovers every tool, then generates attacks from each tool's own schema: missing required
fields, wrong types, empty strings, 100,000-character strings, injection-shaped payloads. Then
it scores what survived.
I have run it against 13 publicly installable MCP servers - including ones maintained by
Anthropic, Google, Microsoft and MongoDB.
What I found:
- Two servers had real input-validation gaps. Both were reported privately to their maintainers,
through the channel each of them asked for. I am not naming them until they have had time to
respond.
- Three more would not start at all on a clean install, including two official reference
servers. Their code was fine. Their dependency pinning was not.
- Five false positives in my own tool. I fixed them and wrote all of it up publicly, because a
scanner that fails well-built servers is worse than no scanner at all. Existing tools in this
space run roughly a 78% false-positive rate.
That last one is the point. Anyone can run a scanner and send you a list of scary-looking
findings. Knowing which of them are real is the actual skill, and I have shown my working in
public: github/junaidshahid-dev/mcp-probe
Why this matters for your project: MCP tools are called by AI agents, and agents produce
malformed arguments constantly. That is the normal case, not the edge case. A tool that trusts
its inputs will break in production, and it will break in a way that is very hard to trace back.
I build servers that validate properly, and I hand you the audit report that proves it.
How I work:
- I ask specific questions before writing code, so you don't pay for a wrong guess
- Tests and documentation come with the job, not as an upsell
- You get the full source, and it runs on your infrastructure, not mine
- If I think your idea has a problem, I tell you before you pay me, not after
Computer Science graduate, FAST-NUCES. Based in Lahore, and I work across US and EU hours.
Tell me what you're building or what's breaking, and I'll tell you honestly whether I'm the
right person for it.
Steps for completing your project
After purchasing the project, send requirements so Junaid can start the project.
Delivery time starts when Junaid receives requirements from you.
Junaid works on your project following the steps below.
Revisions may occur after the delivery date.
Agree the tool list before any code
I turn your description into a concrete list of tools, each with its inputs and outputs, and send it for approval. Cheapest possible place to change your mind is here, before anything is built.
Build the server with validated inputs
Python and FastMCP, with typed schemas on every tool so malformed arguments are rejected cleanly instead of crashing the server or silently doing the wrong thing.