You will get an MCP server that connects your Python code, data or simulation to Claude
Project details
I build MCP servers — the layer that lets an AI assistant call your Python code as real, typed tools instead of guessing at it in prose.
I have shipped two for my own scientific software. One exposes 78 tools over an optics-simulation add-on and is published to PyPI with a release workflow behind it. The other exposes nine tools over a materials and electrochemistry simulator under Apache-2.0 — still pre-alpha, and it says so in its own manifest, because a tool that hides its limits is worse than no tool. So the parts that usually go wrong are ones I have already had to solve: argument validation, error messages a model can recover from, tools that stay honest about what they do not know, and packaging so your client installs it in one line.
What you get: a working server; every tool with typed arguments and a docstring written for the model rather than for a human skimming; a pytest suite that exercises the tools without a live model; and a README with the exact config block to paste into your client.
Send me the functions or the repo. If MCP is the wrong tool for what you want, I will say so before you buy.
I have shipped two for my own scientific software. One exposes 78 tools over an optics-simulation add-on and is published to PyPI with a release workflow behind it. The other exposes nine tools over a materials and electrochemistry simulator under Apache-2.0 — still pre-alpha, and it says so in its own manifest, because a tool that hides its limits is worse than no tool. So the parts that usually go wrong are ones I have already had to solve: argument validation, error messages a model can recover from, tools that stay honest about what they do not know, and packaging so your client installs it in one line.
What you get: a working server; every tool with typed arguments and a docstring written for the model rather than for a human skimming; a pytest suite that exercises the tools without a live model; and a README with the exact config block to paste into your client.
Send me the functions or the repo. If MCP is the wrong tool for what you want, I will say so before you buy.
AI Algorithms
Large Language Model, Transformer ModelAI Applications
Conversational AIAI Development Language
PythonAI Models
GPT-4What's included
| Service Tiers |
Starter
$120
|
Standard
$280
|
Advanced
$550
|
|---|---|---|---|
| Delivery Time | 3 days | 7 days | 14 days |
Number of Revisions | 1 | 2 | 3 |
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 |
Frequently asked questions
About Emir
Physics Simulation & Scientific Imaging in Python | Blender, Optics
Istanbul, Turkey - 3:50 pm local time
OPTICAL SIMULATION - blender-optics-simulator (GPL-3.0, 15 stars)
A Blender add-on that traces real optical benches: live beam tracing, kinematic-mount auto-alignment, Jones/Stokes polarization, interference, Gaussian beams. Each phenomenon self-checks against a textbook closed form, and a structural mesh check runs in CI on every push. A companion MCP server on PyPI (blender-optics-mcp) lets an AI agent build, align and render a bench. Concept DOI 10.5281/zenodo.20778997.
A researcher in ultrafast spectroscopy uses it to preview optical setups before building them in the lab. Their requests drove optical-path-length statistics, a shutter element, unit-scale awareness and square apertures.
DEVICE PHYSICS - tarhan (Apache-2.0)
A validation-first 0D/1D materials simulator: finite volume, Scharfetter-Gummel, damped Newton, now extended to 2D. Every physics claim is pinned to a source-cited reproduction test. Cross-checked against DEVSIM 2.10.0, an independent production TCAD code: built-in potential within 0.5730 microvolts, current ratios within 8.4e-5 without recombination and 2.7e-3 with SRH on. DOI 10.5281/zenodo.21761218.
QUANTITATIVE IMAGING - dhm-hybrid (Apache-2.0)
Digital holographic microscopy: off-axis reconstruction, quantitative phase imaging, autofocus, and reference-free phase retrieval through a hybrid classical + CNN pipeline. Built alone from fifteen months of lab work. The default focus algorithm is pinned by a nine-scene real-lab benchmark rather than by preference.
WHAT I CAN DO FOR YOU
- Optical and device simulation, with the assumptions written down
- Scientific Python: NumPy/SciPy, PyTorch, PySide6/Qt desktop tools, CI, real test suites
- Image processing: FFT and spectral methods, phase retrieval, registration, measurement
- MCP servers and automation over scientific tools
- Physics animation in Blender and Manim where the motion comes from a solve, not keyframes. Four years as a video editor and sound designer at Deus Studio, so you get a finished cut rather than raw renders.
HOW I WORK
I say what I will verify before I build it, I test against ground truth or an independent code, and I tell you when a number is not verified and why. If a brief needs light, optics or device physics to be correct rather than merely plausible, that is the part I do.
For the record: I am a physics undergraduate at Gebze Technical University with about three semesters left, not a senior engineer. The case is the code, and it is all linked above.
Tell me about your problem and I will reply with a concrete plan, not a template.
Steps for completing your project
After purchasing the project, send requirements so Emir can start the project.
Delivery time starts when Emir receives requirements from you.
Emir works on your project following the steps below.
Revisions may occur after the delivery date.
Agree the tool list
I read your code and send back the tools I propose, with their arguments, before writing anything. Cheaper to argue here than after.
Write the server and the tests
Typed arguments, docstrings the model reads, and a pytest suite that calls every tool without a live model. You see the test output, not just my word.