For Coding Agents

Say

“Deploy it with LiteBin.”

AI Agent:Consider it shipped — live at portfolio.l8b.in.

Your coding agent gets deploy, logs, health checks, and runtime secrets as first-class MCP tools. No dashboard visits, no memorized commands, no pasted passwords.

MCP-native Scoped tokens JSON-first Write-only secrets

agents that hit l8bin.com/llms.txt discover the tools and can install the MCP server themselves

01 · Setup

Zero to agent-driven.

1

Setup LiteBin

once, on any VPS with Docker

Already running LiteBin? Skip ahead, this step isn't for you.

curl -fsSL https://l8b.in | bash

Fresh install? Create the first admin account when the installer finishes (the quickstart walks the prompts).

2

Set up litebin mcp

your agent
setup litebin mcp from l8bin.com

The agent takes it from here. It writes .mcp.json with the npx shim (no CLI install, the binary downloads on first use), then has you reload the client so it can call the setup tool: it asks for your server URL, runs the pairing, and hands you an approval link. Open it, pick the scope, approve.

rather register it yourself?

claude mcp add litebin -- npx -y l8bin-mcp

After the reload, ask your agent to run the setup tool; it pairs auth from there.

3

Reload the client

Your agent already registered the MCP server in .mcp.json; one reload and the tools are live.

  • ✔Claude Code, Cursor and ZCode pick up the .mcp.json your agent wrote on next open; approve it when prompted
  • ✔Registered manually (Codex etc.)? Just open the client; the tools are already live

that's the whole setup · from here it's conversation ↓

02 · In action

Talk. It ships.

deploy → verify → crash → fix · no human in the loop

claude code · ~/portfolio

03 · Toolbox

The toolbox.

15 tools, one stdio server, no extra daemon

setup deploy status list logs env_list env_push url stop start restart domain_set domain_remove delete doctor

works with Claude Code, Codex, Cursor, ZCode, Copilot & any MCP client

04 · Bounded

Bounded by design.

you approve the agent, it never gets the keys

Scoped

read → deploy → manage → admin, picked at pairing

Project-bound

token can be locked to one project, blind to the rest

Write-only secrets

agents can set env values, never read them back

Revocable

one click in Settings → Access Tokens, all actions audited

For your agent.

everything it needs, by URL

llms.txt

the guide: setup recipe, API patterns, CLI commands

llms-full.txt

llms.txt + OpenAPI spec + full CLI reference, one fetch

npm package

the MCP server, no pre-installed CLI needed

AGENTS.md template

teach agents that don't speak MCP the workflow

docs/agents.md →