start in one command

there is one command. it does the whole setup.

npx innernet

it opens a browser to sign you in, then finds every ai tool on this machine that speaks mcp — claude code, cursor, codex, windsurf, gemini cli, claude desktop, vs code, cline, zed, continue — and writes the one line each of them needs. run it inside a repository and that folder gets a memory of its own; netti reads the readme and the file tree and shapes the memory around what the project actually is.

what it wrote

every client is pointed at the same door:

https://innernet.live/api/mcp

that is the entire configuration. on first use the client opens a browser sign-in — you approve once at innernet.live and it holds a scoped token from then on. you never paste a key into a chat, and you shouldn't: if an ai ever asks you to, hand it the url instead and let the sign-in run.

tools that can only spawn a command rather than call a url (claude desktop, codex, zed) are given innernet mcp, a thin bridge to the same endpoint.

the anchor in the repo

inside a repository, npx innernet commits a small .innernet/connection.json. it records the project slug and nothing secret. because it is committed, the link travels with the code — someone you've shared the memory with clones, runs the same command, and inherits the same context.

if you'd rather wire one client by hand

claude code is a one-liner:

claude mcp add --transport http innernet https://innernet.live/api/mcp

cursor, cline, continue and zed take the url under their mcp settings and run the same sign-in. the per-client walkthroughs live in the docs under connect an ai.

how to know it worked

open a fresh session in any connected tool and say load my projects from innernet. it should list what you have. nothing there yet? say save this conversation to a new innernet project and watch it appear on your dashboard.