floom

The production layer for AI agents. Deploy Python scripts as cloud automations with a shareable web UI, REST API, and MCP endpoint. No Docker, no infra, no YAML.

Tell your AI agent "deploy on floom" and your script is live in seconds.

What floom does

You write a Python script. floom turns it into:

• A web UI with typed inputs and outputs anyone can use
• A REST API other services can call
• An MCP endpoint AI agents can discover
• Managed secrets, version history, and scheduling

No Docker. No CI/CD. No infrastructure. Just Python in, production out.

Who it's for

• Developers who build Python automations and need to share them with non-technical people
• AI agencies deploying tools for clients
• Anyone using Claude Code or Cursor who wants to go from localhost to production instantly

Quick start

Install the Claude Code skill

git clone https://github.com/floomhq/floom.git ~/tmp/floom && ~/tmp/floom/scripts/setup

Deploy your first script

1. Type /floom in Claude Code
2. Point it at any Python script
3. It adapts, tests in a sandbox, and deploys
4. You get a live URL to share

Or tell any MCP-capable agent: "deploy this on floom"

Write a script from scratch

import os
from google import genai

def run(url: str) -> dict:
    client = genai.Client(api_key=os.environ["GEMINI_API_KEY"])
    response = client.models.generate_content(
        model="gemini-2.0-flash",
        contents=f"Summarize the company at {url} in 3 sentences."
    )
    return {"summary": response.text}

That's it. floom handles the rest: sandbox testing, deployment, UI generation, secret injection.

How it works

1. Write a Python script with a run() function
2. Deploy via Claude Code (/floom), Cursor, or any MCP-capable agent
3. Test runs automatically in an E2B sandbox (EU-hosted, SOC 2 certified)
4. Share the live URL with anyone

See BUILDING.md for the full protocol: manifest format, input/output types, API reference.

API

All endpoints at dashboard.floom.dev/api/. Full reference in BUILDING.md.

# Upload code
curl -X POST https://dashboard.floom.dev/api/artifacts \
  -H "Authorization: Bearer $API_KEY" \
  -d '{"code": "def run(x): return {\"result\": x}", "manifest": {...}}'

# Test in sandbox
curl -X POST https://dashboard.floom.dev/api/test \
  -d '{"artifactId": "...", "inputs": {"x": "hello"}}'

# Deploy
curl -X POST https://dashboard.floom.dev/api/deploy \
  -d '{"artifactId": "..."}'

Features

Feature	Status
Deploy from any AI agent	Available
Web UI with typed inputs/outputs	Available
REST API per automation	Available
Managed secrets	Available
Version history and rollback	Available
Sandbox testing (E2B)	Available
Scheduling (cron)	Available
File uploads (PDF, CSV, XLSX)	Available
MCP endpoint	Available (@floomhq/mcp-server)
Multi-file projects	Coming soon
Workspaces and teams	Planned

Compared to

	floom	n8n	Vercel	Modal
Deploy from AI agent	Yes	No	No	No
No-signup first deploy	Yes	No	No	No
Auto-generated UI	Yes	No	No	No
Python-native	Yes	Partial	No	Yes
Sandbox testing	Yes	No	No	No
10-second deploy	Yes	No	Partial	No

Stack

• Runtime: E2B sandboxes (EU-hosted, SOC 2 Type II)
• Backend: Convex
• Auth: Clerk
• Storage: Cloudflare R2
• Frontend: Next.js on Vercel

Links

• Dashboard: dashboard.floom.dev
• Docs: docs.floom.dev
• Protocol reference: BUILDING.md

FAQ

What is floom?

floom is the production layer for AI agents. You write a Python script, and floom turns it into a cloud automation with a shareable web UI, REST API, and MCP endpoint. No Docker, no infra, no YAML. Tell your AI agent "deploy on floom" and your script is live in seconds.

How is floom different from n8n, Zapier, or Make?

n8n, Zapier, and Make are visual workflow builders where you connect nodes in a GUI. floom is code-first: you write a Python function, and the platform generates the UI, API, and infrastructure automatically. There is no drag-and-drop editor. You deploy from your terminal or AI agent, and non-technical teammates use the generated web UI to run it.

How is floom different from Vercel or Modal?

Vercel deploys full web apps. Modal deploys Python functions but requires their SDK and decorators. floom deploys a single Python run() function and automatically generates a web UI with typed inputs and outputs, so anyone on your team can use it without writing code or calling an API.

Does floom work with Claude Code?

Yes. Install the skill with git clone https://github.com/floomhq/floom.git ~/tmp/floom && ~/tmp/floom/scripts/setup, then type /floom in Claude Code. It adapts your script, tests it in a sandbox, and deploys it.

Does floom work with Cursor, Windsurf, or other AI agents?

Yes. Any MCP-capable agent can deploy to floom. The protocol is a standard REST API: upload code + manifest, test, deploy. See BUILDING.md for the full API reference.

Is my data safe?

Code runs in E2B sandboxes, which are EU-hosted and SOC 2 Type II certified. Each run gets an isolated sandbox that is destroyed after execution. Secrets are encrypted at rest and injected at runtime, never exposed in logs or UI.

How much does it cost?

floom is free to start. See dashboard.floom.dev for current pricing details.

What languages does floom support?

Python. Your script needs a run() function that takes typed parameters and returns a dict. Dependencies are pip-installed automatically from the manifest.

Can I schedule automations?

Yes. Add a schedule field (cron syntax) to your manifest. For example, "schedule": "0 9 * * 1" runs every Monday at 9am. You can also set scheduleInputs with default values for scheduled runs.

Can I upload files (PDF, CSV, Excel)?

Yes. Use the file input type in your manifest. The user uploads a file through the web UI, and your run() function receives an R2 URL string. Download it with requests.get() or httpx.get() inside your function.

License

Open source. See LICENSE for details.