Service Buddy
Tell Claude what your app needs in plain English — user accounts, file uploads, background jobs, and 10 more — and it sets them up for real on your own machine, usually in under a minute. It only tells you something is ready once it has actually checked.
This is what you'll actually see
No new tool to learn, no form to fill out. You type into the chat window you already use — Claude Code, Cursor, or Claude Desktop — the way you'd ask a person. Everything runs on Floci, a free emulator of AWS that runs in Docker on your own machine, so nothing touches a real cloud account or costs money. Below is a recorded conversation, replayed exactly as it happened.
When someone uploads a photo, resize it, then send them a confirmation email.
What it can set up today
These are the 13 things Service Buddy can set up today. Each one comes with a tested recipe and its own check, shown on the card, which has to pass before you're told it works. Ask for something that isn't on this list and it tells you so rather than improvising.
Missing something you need? Request a capability.
The numbers, not the pitch
Every AI tool says it's fast and easy to talk to. Here's the actual difference it makes to what you, the founder, see land back in your chat window for the exact same request:
When someone uploads a photo, resize it, then send them a confirmation email.
“Done — created an S3 bucket, wired a Lambda function to resize on upload, and set up SES for the confirmation email.”
The step count and time are measured from this exact request run on an agent with no Service Buddy. The reply wording is illustrative, built from the jargon (S3, Lambda, SES) those runs actually used.
When someone uploads a photo, resize it, then send them a confirmation email.
“Done and verified — photos people upload get resized, and each person gets a confirmation email.”
Summarises the recorded run replayed at the top of this page: three pieces, each re-checked before the reply.
We didn't just eyeball that difference. We ran a fixed set of 6 founder requests, one run each, with and without Service Buddy, against real local infrastructure, and measured what happened.
A small benchmark (6 requests, one run each), so read it as a direction, not a guarantee. The full benchmark writeup has every number, honest misses included.
Numbers from a fixed benchmark are one kind of proof. Here's another: the three most common requests, set up and then re-checked from scratch against a real local engine. Recorded on 2026-09-25.
This times Service Buddy's own set-up and re-check speed against a real local Floci instance. It measures something different from the benchmark above and doesn't replace it. Raw log.
Ready to use it on your own idea?
Up and running in a few minutes, on your own machine.
You need: macOS, Linux, or Windows via WSL · Docker running (setup can install Colima for you) · Python 3.11+ · the AWS CLI (used locally, no AWS account needed) · git.
Clone and run setup
Run the command below once. It checks for Docker and Floci, offers to install what's missing (it asks first), creates its own Python environment, and connects itself to Claude Code and Cursor.
Open your chat app in that folder
Start Claude Code (or open the folder in Cursor) inside service-buddy
and approve the server when asked.
Describe what you want
Try “Let people sign up and upload a profile photo.” No commands to memorize after that.
Run setup once
Claude Desktop still needs the local engine, so run the command below once in a terminal. Answer no when it offers to connect Claude Desktop; the bundle in the next step does that.
Install the bundle
Download service-buddy.mcpb from the
latest release, double-click it, and click Install.
Describe what you want
Open a new chat in Desktop's Chat tab (the Code tab doesn't load extensions) and ask in plain English. Service Buddy sets things up and checks them itself; there's nothing to run.
git clone https://github.com/ahmedkeewan/service-buddy.git && cd service-buddy && ./setup.sh
The README has the full setup guide, troubleshooting, and how to uninstall.
There's more once you're in
Chat is the front door, not the whole house. Three things worth knowing about once you're actually using it:
Watch it work, in a browser too
The conversation at the top of this page is a replay of a real run. Run
make board and you get a live status page at
localhost:7777 that updates in your browser as things get set up, side by
side with the chat. You don't need it to get started.
Ask what it costs. Get a real answer.
Ask what something would cost on real AWS and it pulls live numbers from AWS's public price list instead of guessing. On 2026-09-25, 5 GB of file storage priced out at $0.11/month. 6 of the 13 capabilities get live prices; the rest show a clearly labelled estimate.
Setup that asks before installing anything
./setup.sh checks for Docker and Floci, offers to install whatever is
missing (only with your yes), and connects Service Buddy to Claude Code, Cursor, and
Claude Desktop in one pass. Safe to run again any time. macOS, Linux, and Windows via
WSL.
For engineers
Service Buddy is an open-source MCP server (MIT licensed). The harness README covers the architecture and tools, AGENTS.md is for AI coding agents connecting to it, and the repository has the design decisions and benchmark data.
Design patterns worth stealing: shared-backend naming-scope isolation · flock-based state locking