The MCP server with everything in it.
One MCP server gives AI coding agents the tools to ship production software and the people who hold it decide what it learns next.
The engine
One server. One toolbox. Anything you can describe.
You never leave your own tools. putty comes to you.
Plug in
One line of config — paste putty’s MCP server into Claude Code, Cursor, Hermes, or point your service at the API with a key. Install time: one paste. Nothing new to learn; you prompt like you already do.
Describe
Say it like you would say it to a friend. “A booking page, calm colors, works on phones.” No spec doc. No ticket. No spiral. Behind the curtain, the engine takes over.
The mold
Jev routes the request to the right corner of the toolbox — the hardened template closest to your spec, the tools and skills to finish it. The LLM writes; Jev verifies every step with typed decisions: right template, right parts, quality checks passed. Production-grade parts — not glue code you babysit for three weekends. Not code that compiles eventually — code that ships.
Ship
Deploy anywhere. Your apps are yours — nobody holds them hostage here.
- describe
- route
- retrieve
- mold
- verify
- ship
The toolbox
Deep where it matters.
Tools for the jobs agents actually fumble — deploying, wiring, testing, shipping — not a museum of party tricks.
Finished, not starting points.
Templates that are hardened, tested, production applications — molded to your spec, never static files you fix alone.
A catalog with a heartbeat.
New tools and templates drop every week, voted in by the people who hold the engine. It never sits still, so neither do you.
For the vibecoders
You don't need to know how to code. You need to know what you want. putty handles the distance between the two.
For the developers
Stop rebuilding the same toolset on every project. One key, whole toolbox. MCP for your agent, a plain API key for everything else.
For the companies
The template market sold static files for twenty years and called it a product. putty's catalog is living, community-directed, and molded to the request by the agent using it. Thousands of hours a year, handed back.
Today the engine travels to your tools. It was built to move.
What you can create
If you can describe it, it’s on the table.
The catalog is deep and growing every week, because the community votes in new skills constantly. A taste:
The Brain
One model writes. One model decides. Every mold runs a two-model architecture. The LLM writes — code, copy, structure. Jev decides — a System One decision model making the micro-calls inside every build (routing, verification, classification, quality checks) as typed decisions with calibrated probabilities. Your build doesn’t just get answers; it gets answers and how sure to be.
Most builders run every decision through an expensive chat model. putty runs them through infrastructure built for decisions.
A chat model is a poet. A decision model is a referee. putty fields both.
Under the clay
MCP-native
Any agent that speaks Model Context Protocol plugs straight in: Claude Code, Cursor, Hermes, your own harness.
REST + API keys
Services, pipelines, and cron jobs call the engine like any other infrastructure — the decision layer included, so your pipelines get calibrated probabilities, not vibes.
Hardened templates
Versioned, tested, production applications, molded per request — never static files.
Decision-grade orchestration
Every build runs through Jev's typed decision loops: routed, verified, classified, quality-checked at machine speed.
Growth Phase
- 01
- 02
- 03
- 04
- 05
- 06
Fair launched. Community-owned from block one.
No presale. No insiders. No roadmap theater — the engine speaks for itself.