One method.
Experts that keep learning.
loop0 reaches your coding agent through MCP servers. The method gives it a way of working. Next to it, a library of experts: what we learn on real projects about one job, kept current, with a stated level of expertise, served to your agent when it needs it.
A way of working for every project
A written foundation, missions you approve, loops checked in real use, reports that are verified. Your agent calls this server to render each mission and to check each report.
https://agmgd-on-loop0-22c0a8ea.fly.dev/mcp
One expert per job
Deploying on Fly.io, putting a domain in front of an application with Cloudflare, making a voice-over with ElevenLabs: each job gets its expert. Your agent loads the ones your project needs through the loop0 address, and an expert that requires another brings it along.
The expert brings the know-how. The token stays with your agent and gives it the hands. loop0 never sees a secret.
The right piece at the right moment
Your agent describes a symptom, such as “files gone after a restart”. It gets the cause, the fix and how to prove it.
A job step by step, such as a first deployment. Each step comes with a check.
What it takes to choose: the products and their differences, the kinds of token, the prices with their date.
What to confirm before a risky action, such as destroying a volume or adding to the bill.
A script or a configuration your agent saves in your project and runs there, such as a relay to publish, with how to use it.
An expert learns from real work.
A file of skills is written once, then it ages. An expert has a version. It changes when the tools change, when our work meets something new, and when an agent using it finds what it did not know.
Everything is written down
We log all our work in a second brain: every deployment, every incident, every fix, as it happens. That record is where an expert’s lessons come from.
A curator, then a person
A curator gathers the new challenges and observations, removes duplicates, checks each lesson and takes out what belongs to one project or one client. A person reviews the change and merges it.
The servers follow the repository
The library is a repository with versions. Each lesson is dated, tested and sourced. The servers serve what the repository says, and nothing is written through them.
One expert per job
Putting a domain whose zone is on Cloudflare in front of an application hosted elsewhere, through Cloudflare's proxy and a relay Worker, with a token the agent holds itself — the token and its permissions, DNS records, zone settings, the edge certificate, what Cloudflare adds or blocks without saying so, the cost in response time, checking from outside and rolling back.
preview · v0.2.0Making a voice-over, music and sound effects with the ElevenLabs API from a project, with a key the agent holds itself — choosing models, voices and formats, estimating credits before acting, keeping what was made, checking the right to publish, and going on without a key.
preview · v0.2.0Deploying and running applications on Fly.io with a token the agent holds itself — products, deployments, images, persistent disks, ports, private access, cost, performance and reporting.
preview · v0.2.2Making a true, finished video of something that runs on a screen — an HTML slide deck, a web page or app, a Godot app — with narration, music and subtitles; from the message and the approved words, through a timed shot list, capture in a headless browser or Godot's Movie Maker, assembly and checks with FFmpeg, to a master, a web version and subtitle files, with a record of where every picture and sound comes from.
preview · v0.2.0Managing a Discord server.
plannedCloudflare, ElevenLabs, Fly.io and Video production are previews: every piece is dated and sourced, and says whether we measured it or read it at the vendor. None has been used end to end by an agent on live work yet; they go online now so that real use improves them.
How you use one
- You connect loop0 to your agent once, with one line. Nothing else to configure for a new expert.
- You ask for an expert by name. Your agent loads it, and the experts it requires come with it in the same answer.
- Your agent searches from a symptom or a job, and reads one piece in full when it needs it.
- You give your agent the token the expert asks for, in your own project. Your agent does the work and tells you which pieces it followed.
claude mcp add --transport http loop0 https://agmgd-on-loop0-22c0a8ea.fly.dev/mcp
Then, in your agent: “load the fly-io expert”. An agent that needs one expert and nothing else can connect its own address instead: https://agmgd-on-loop0-22c0a8ea.fly.dev/mcp/fly-io
What stays true
- No AI on our side: an expert is written knowledge, and your agent does the thinking and the work.
- No account, nothing of yours stored: a call is answered and its content forgotten. We only count that it happened, in a public log.
- No secret reaches us. The token stays in your agent’s session, and no tool of an expert takes one.
- Read-only servers: a lesson enters the library through review, never through a request.
- Nothing is called proven that is not: every piece says whether we measured it or read it in the vendor’s documentation.
FAQ
Is an expert an AI agent?
No. An expert is written knowledge, served by pieces. Your own agent reads it and becomes the specialist: it does the work, with its own tools and the token you gave it. loop0 never calls a model.
Where does the token go?
It stays with your agent, in your project, encrypted. The expert tells your agent which token a job needs and how to use it without ever showing it. loop0 never receives it: no tool of an expert takes a token.
How is this different from a skills file?
A skills file is one block of text that someone wrote once. An expert is served by pieces, at the moment your agent needs one, and it has versions: every lesson is dated, tested and sourced, and the library changes as the work goes on.
Where does the knowledge come from?
From our own work on real projects, logged as it happens, then curated and reviewed by a person. Nothing confidential from a project or a client is kept.
Do I have to connect each expert?
No. Connect loop0 once: its address lists the experts and loads one by name, with the experts it requires. Each expert also keeps an address of its own, for an agent that needs only that one.
Can I use an expert without the loop0 method?
Yes. Each expert has its own endpoint. The method and the experts work together, and each works alone.
What does “preview” mean?
Every piece of the expert is dated and sourced, and every command or API call it gives is checked against the real tool or the vendor’s own description. What we have not done yet is run the whole expert end to end on live work. Until then it says preview, and each piece says whether we measured it or took it from the vendor’s documentation.
Start with the method today
The experts plug into the same agent, next to it.