nnesly
Menu
TRSign in

Tell your AI assistant to deploy — it can. The whole platform is a set of tools your agent already knows how to call.

nesly is built for the developer who writes code with an AI assistant. Every command speaks --json with stable exit codes and --yes for no prompts; the same binary is an MCP server for Claude Code and Cursor, and a hosted MCP endpoint with OAuth serves ChatGPT and Claude.ai. When a deploy fails, nesly diagnose names the failure and the fix in a sentence an agent can act on.

Get startedRead the documentation
chat
You:    Deploy this to nesly, add a Postgres database and tell me the URL.

Agent:  Running nesly deploy --preview --yes --json …
        Added postgres (DATABASE_URL injected).
        Preview is live: https://my-app-x7k2-pr.nesly.net
        The health check passed. Promote it to production?

You:    Yes, and put it on shop.example.com.

Agent:  Promoted. Set this record at your DNS:
          shop  CNAME  my-app-x7k2.nesly.net
        I will verify once it resolves.

How it works

  1. 1

    Connect your assistant

    Claude Code: /plugin marketplace add neslyhq/claude-plugin. Cursor or Claude Code without the plugin: nesly mcp setup. ChatGPT or Claude.ai: add the hosted MCP endpoint and sign in once.

  2. 2

    Ask in plain words

    "Deploy this", "show me the logs of the last failure", "add a database", "roll back". The agent calls the same tools the CLI exposes; previews come first unless you ask for production.

  3. 3

    Review the result

    The agent hands back a URL, not a stack trace. If something fails, diagnose tells it what to change in your code, and it does.

What you get

MCP, three ways

A local stdio server for Claude Code and Cursor, a Claude Code plugin with deploy, debug and rollback skills, and a hosted Streamable HTTP endpoint with OAuth for ChatGPT and Claude.ai.

Every command is machine-readable

--json on every command, stable exit codes, --yes for unattended runs. Errors say what to change, not just that something broke.

Diagnosis an agent can act on

nesly diagnose reads the failed deployment and answers with the cause and the fix: a missing start command, a wrong port, a package that did not install.

Preview-first by default

Agents deploy to a preview address and ask before touching production. A bad guess costs a preview, never your live site.

The whole product, not a subset

Projects, deploys, logs, env, domains, DNS, add-ons, AI models, backups, usage and spend caps are all tools. What you can do, your agent can do.

Documentation agents can read

Every docs page has a Markdown twin and the site publishes llms.txt, so your assistant reads the same docs you do.

Pricing

The CLI, the MCP servers and the Claude Code plugin are free on every plan. Your agent uses your account's allowances, nothing more.

  • Agent-driven deploys are deploys: build minutes and traffic count exactly as they would from your keyboard.
See plans and allowances

Use cases

Ship from the chat you are already in

You describe the feature, the assistant writes it, deploys a preview, and sends you the URL to try.

Debug a production error at 2 a.m.

"Why is /checkout failing?" — the agent pulls the request logs, reads the diagnosis, patches the code and deploys a preview for you to confirm.

Keep an eye on what it did

Every deploy, domain change and backup an agent makes is in the account's activity log, with nesly activity, exactly like one made by hand.

Questions, answered

Which assistants work?
Claude Code and Cursor over local MCP, Claude Code also through the plugin, ChatGPT and Claude.ai through the hosted MCP endpoint with OAuth. Anything that speaks MCP or runs a shell command with --json works.
Can an agent break production?
Agents deploy previews first and ask before promoting. Production changes only when a release answers its health check, and nesly rollback undoes it in one step.
Does the agent get my credentials?
Local MCP shares the CLI session you signed into; the hosted endpoint signs you in once with OAuth. No key is pasted into a chat.
Is there a cost for the MCP endpoint?
No. Plans and allowances are the same whether a command comes from you or from your assistant.

Related products

Applications

Push your code; it is live in minutes. Every change gets its own address, and one command puts the last release back.

AI models

An open model your app calls with the OpenAI SDK it already uses. One command adds it; you pay per token, in your currency.

Databases and add-ons

PostgreSQL, MySQL, SQL Server, MongoDB, Valkey, object storage and outgoing mail — each attached with one command, connected by environment variables.

Let your assistant deploy the next one

Install the plugin or run nesly mcp setup, then say "deploy this".

Get startedRead the documentation