Agent Lab Deploy · MCP server

Deploy from Claude Code.
One line to connect.

Agent Lab Deploy is an MCP server. Connect it to Claude Code or Codex once, sign in in your browser, then say "deploy this": the app is created, uploaded, built and given a preview URL, and goes live only when you say so. No API key to paste.

Start free Setup docs

Claude Code · agentlab-acme

> deploy this folder to acme

sites_createquote-calculator → quote--acme.hosting.agentlab.run

sites_deployrelease 1 · upload address issued

sites_deploy_finalizebuild queued

sites_deploy_statusbuilt · secret scan passed

Release 1 is built. Preview: r1--quote--acme.hosting.agentlab.run. Make it live?

Connect

One command per tool.

Replace acme with your organization's name. The console's MCP page shows it filled in.

CLAUDE CODE

Claude Code

claude mcp add --transport http agentlab-acme https://mcp.agentlab.cresclab.com/mcp

A browser tab opens to sign in and pick the organization. Done.

CODEX

Codex

codex mcp add agentlab-acme --url https://mcp.agentlab.cresclab.com/mcp
codex mcp login agentlab-acme

The second line signs you in. Codex keeps the connection in its config.

OTHER CLIENTS

Any MCP client

https://mcp.agentlab.cresclab.com/mcp

Add it as an HTTP (streamable) MCP server. It needs OAuth 2.1 with PKCE and Client ID Metadata Documents; a client that can only register itself dynamically can't connect yet.

Already signed up? The console's MCP page has these commands with your organization filled in, and a prompt you can paste into your tool.

From chat to URL

Connect → Say it → Preview → Live

The agent follows a published guide, and asks you before anything goes live or starts billing.

Connect
One command, one browser sign-in. The connection is yours: it can do what you can do in the console, nothing more.
Say it
Tell your tool to read app.agentlab.cresclab.com/agent/deploy.md and deploy. It creates the app, uploads the zip straight to the platform and starts the build.
Preview
Every release gets its own URL. The tool reports the build result and the secret scan, and waits.
Live
Say which release goes live. Rolling back is the same call with an older release.

35 tools

Everything the console does, as tools.

Grouped the way the docs list them. Every tool takes the organization, and names the branch when there is more than one.

15 tools

Apps and releases

sites_create, sites_deploy, sites_deploy_finalize
sites_deploy_status, sites_deploy_retry, sites_files
sites_promote goes live and rolls back
sites_build_plan shows how a source will build, before uploading; sites_build_settings_get, sites_build_settings_update
sites_runtime_status, sites_restart, sites_lifecycle, sites_purge

17 tools

Databases and jobs

databases_create (PostgreSQL 18, preview first, then confirm)
databases_connect_app writes the connection into an app's variables; the password is never shown
databases_allowlist_update, databases_stop, databases_start, …
scheduled_jobs_create, scheduled_jobs_run, scheduled_jobs_runs, …

3 tools

Workspace

environments_list: the organization's branches
usage_summary: this month's usage and cost
docs_read: these docs as Markdown, so the agent looks things up itself

Every tool, with what it returns →

Why this one

An MCP server that behaves like a colleague with your badge.

Vercel and Netlify have MCP servers too. Here is what Agent Lab's does differently.

Your permissions, not a pasted token

Sign-in is OAuth in the browser. The agent acts as you, inside your organization, and an admin can revoke the connection any time.

Going live is a call you approve

Building a release and making it live are separate tools. Every release keeps its own URL, so rollback is one call away.

Checked before it builds

Leaked API keys, bundled .env or .git folders and sensitive files stop the upload, and the failure tells the agent exactly what to fix.

Runs in Taiwan

Apps and PostgreSQL databases run on Google Cloud in Taiwan and belong to your organization.

FAQ

Before you connect

What is the Agent Lab MCP server?

It is the deploy platform of Crescendo Agent Lab exposed as an MCP (Model Context Protocol) server at https://mcp.agentlab.cresclab.com/mcp. AI coding tools such as Claude Code and Codex connect to it and deploy websites, container apps, agent services and MCP servers from the chat, with a preview URL and rollback for every release, PostgreSQL databases and scheduled jobs.

Do I need an API key?

No. Sign-in is OAuth 2.1 in your browser. The connection carries your own permissions in your organization and can do exactly what you can do in the console. There is no key to paste into a config file, and nothing to rotate.

Which AI tools work?

Claude Code and Codex connect with one command each. Any other MCP client that supports OAuth 2.1 with PKCE and Client ID Metadata Documents can add the same HTTP endpoint. A client that can only register itself dynamically cannot connect yet. Setup steps →

How does the agent know the procedure?

Tell it to read https://app.agentlab.cresclab.com/agent/deploy.md, a step-by-step guide written for agents: confirm the address label, upload, wait for the build, ask before going live. It can also call docs_read to look anything up in these docs.

Is it free?

The MCP server is part of Agent Lab Deploy, so your plan decides what you can deploy: the Free plan runs 3 static websites, and apps, databases and scheduled jobs start on Pro at $29 a month before tax. Deploy pricing →

Deploy from Claude Code. One line to connect.