Rivtor
Rivtor helps developers deploy full-stack applications using a CLI or an MCP-compatible AI coding agent.
Dokümantasyon
Full-stack apps.Fully live.
Deploy your full-stack application: frontend, backend, Postgres, Redis and storage from one call by your AI agent, or one CLI command. Live in about 2 minutes, all on one private network.
No card. Works with Claude Code, Cursor, Codex and any MCP client.
you › Ship this to production.
● rivtor.deploy(path: "./my-app")
- upload142 files, 3.1 MB packed2.8 s
- detectweb: Next.js 15 · api: FastAPI0.4 s
- planpostgres, redis · api URL → web0.2 s
- provisionp_my_app · redis · rv-my-app1.4 s
- buildweb 58 s ∥ api 41 s58 s
- migratealembic upgrade head7 s
- releasehealth checks passed6 s
{
"status": "live",
"urls": {
"web": "https://my-app.rivtor.app",
"api": "https://my-app-api.rivtor.app"
},
"DATABASE_URL": "postgresql://p_my_app:•••@pg:6432/p_my_app",
"durationMs": 118400
}
claude › Live in 1 min 58 s. Frontend, API and Postgres are wired together.
Three accounts for one app.
A full-stack app today means a frontend host, a backend host and a database provider. Your agent can write the code in minutes, then waits on you to wire three dashboards together.
| Vercel + Render + Supabase | Rivtor | |
|---|---|---|
| Accounts | 3: Vercel, Render, Supabase | 1 |
| Dashboards to watch | 3 | 1, and your agent |
| Environment variables | Copied by hand between dashboards | Wired at deploy: DATABASE_URL, REDIS_URL, API URLs |
| App to database | About 1.8 ms per query, across two providers | < 1 ms on the same private network |
| Getting it live | Sign up three times, connect repos, create a database, paste URLs | deploy(./my-app) |
| When a build fails | A wall of logs for you to read | Cause, file and fix your agent acts on |
Eight steps. About two minutes.
What happens after your agent calls deploy, for a typical Next.js + FastAPI + Postgres app. No human input at any step.
Pack it. Skip the secrets.
The CLI or local MCP server packs the project as one tarball, respecting.gitignore..env files are never uploaded; secrets go through env_set.
2–5 s
deployment dpl_7h2k · my-appbuilding
-
upload142 files packed · 3.1 MB ·.env skipped
-
detectwaiting
-
planwaiting
-
provisionwaiting
-
buildwaiting
-
migratewaiting
-
releasewaiting
-
returnwaiting
-
Upload 2–5 s
-
Detect < 1 s
-
Plan < 1 s
-
Provision < 2 s
-
Build 20 s – 2 min
-
Migrate 5–20 s
-
Release 5–10 s
-
Return ≈ 2 min total
It reads the repo. No config file.
Flip what is in your repo and watch the plan change. Detection works from your files alone; a deploy.toml only overrides what it found.
- There is a Next.js app in /web. Container on port 3000 at my-app.rivtor.app. A static export goes to Cloudflare's edge instead. PORT
- The API in /api is FastAPI.
- The API imports SQLAlchemy. Postgres 17 database p_my_app with its own role, in under 2 s. alembic upgrade head runs before release. DATABASE_URL
- A Celery worker reads REDIS_URL.
- Uploads go through boto3.
plan1 servicePostgres2 env vars wiredready to deploy
Also recognised: Vite, Astro, plain HTML and CRA as static sites on the edge; Express, Fastify, NestJS and Hono; Flask and Django; Go, Rust, Ruby, Java and PHP through Railpack; any Dockerfile; BullMQ and RQ workers. Full detection tables
deploy → failed
{
"status": "failed",
"stage": "release",
"service": "api",
"code": "missing_python_module",
"cause": "Python module 'stripe' not found (imported in payments.py:3)",
"file": "payments.py:3",
"fix": "Add 'stripe' to requirements.txt (or pyproject.toml dependencies) of the 'api' service, then call deploy again",
"retryable": true,
"docs": "https://rivtor.com/docs/errors#missing_python_module"
}
deploy → needs_input
{
"status": "needs_input",
"stage": "plan",
"code": "missing_secrets",
"cause": "The app reads a secret that is not set: STRIPE_SECRET_KEY",
"fix": "Ask the user for STRIPE_SECRET_KEY, call env_set with them, then call deploy again.",
"retryable": true,
"missing": [
{ "name": "STRIPE_SECRET_KEY", "service": "api",
"usedIn": ["api/payments.py:3"] }
]
}
Errors your agent can fix.
Every failure comes back as a structured object: cause first, then the fix. A library of 71 known errors maps raw build output to plain instructions: missing dependency, wrong port, missing secret, failed migration, out of memory. Anything unknown returns the last relevant log lines.
- Agentcalls deploy.
- Rivtorreturns the stage, the cause, the file and the fix. Never a log dump.
- Agentadds stripe to api/requirements.txt.
- Agentcalls deploy again. Only api rebuilds; web reuses its image.
- Rivtorreturns live URLs. The whole loop took under a minute.
Secrets are asked for once, by name and by the line that reads them, and never echoed back. Error reference
Works with your agent.
The MCP server is the product. Paste its URL into your agent, sign in on rivtor.com, then say “deploy this”.
claude mcp add --transport http rivtor https://api.rivtor.com/mcp
# then, inside Claude Code
/mcp
Remote MCP over HTTP. Add it, run /mcp, sign in on rivtor.com. No token to copy.
Fast because it is close.
- Postgres 17: every app gets its own database and its own role, created in under a second.
- PgBouncer in transaction mode, so serverless-style Next.js apps never run out of connections.
- Redis on the project's private network, and an R2 bucket with no egress fees.
- Point-in-time backups: 7 days on Free, up to 30 days on paid plans. Export a dump any time.
- Redeploys in seconds: only the services that changed rebuild. A one-line API change goes live in 3.7 s.
< 1 ms
From your app to its database. Postgres runs on the same private network as your containers, not across the internet.
~1 s
To wake a sleeping free app: 0.58 s for the first API request, 0.96 s for a full page. Render's free tier takes about a minute.
< 30 s
For a static or Vite site, served from Cloudflare's edge.
Measured, not promised.
Same app on four platforms, timed from India. Lower is better.
Redeploy after a one-line changeLower is better
- Rivtor3.7 s
- Vercel + Supabase15.9 s
- Railway62.3 s
- Render67.9 s
App to database, per queryLower is better
- Rivtor0.18 ms
- Vercel + Supabase1.77 ms
- Railway2.69 ms
- Render0.58 ms
Server-rendered page loadLower is better
- Rivtor231 ms
- Vercel + Supabase263 ms
- Railway309 ms
- Render314 ms
Next.js + FastAPI + Postgres, measured from India, 7 Nov 2026. Medians. All results →
Free is for building.
Sign in with GitHub, no card, and get a real full-stack app with a real database. Every new account starts with 14 days of Pro, then settles on Free. Nothing you built has to move when you upgrade.
Free is for personal and non-commercial projects. Once an app earns money, Hobby is $19 a month.
| Projects | 3, each with up to 3 services |
|---|---|
| Static sites | Unlimited, always on, on Cloudflare's edge |
| Container apps | 0.5 vCPU and 512 MB RAM each; sleep after 1 hour idle, wake in about 1 s |
| Postgres | 1 GB per project, never sleeps, 7-day point-in-time backups |
| Redis | 50 MB per project |
| File storage (R2) | 1 GB |
| Bandwidth | 10 GB per month |
| Custom domains | 1 |
| Deploys and MCP tools | Unlimited. Every tool, every framework |
One plan instead of three.
Replaces Vercel Pro + Supabase Pro ($45) and three dashboards. Hobby is $19 for solo builders; Team is $199 with 3 seats. Rupee pricing with UPI in India.
$49per month, Pro
- 3 GB RAM, always on, across unlimited projects
- 20 GB Postgres with 30-day point-in-time backups
- Preview deploys, database branches and priority builds
- 20 custom domains, 100 GB bandwidth, teammates
Questions, answered.
What does one call actually do?
deploy uploads the changed files, detects every service in the repo, creates the Postgres database, Redis and storage it needs, builds everything in parallel, runs migrations, waits for health checks and returns the live URLs and the DATABASE_URL. A second step is only needed when a secret such as an API key is missing.
Do I need a Dockerfile or a config file?
No. Detection reads package.json, requirements.txt, pyproject.toml, go.mod and friends. If a Dockerfile is present it wins. An optional deploy.toml overrides anything detection found, such as a port or a start command.
Which agents can use it?
Any agent that speaks MCP: Claude Code, Cursor, Codex, Windsurf, Zed, Cline or your own. Connect over HTTP at api.rivtor.com/mcp, or run npx -y rivtor mcp locally to upload a folder from disk. Humans use the same API through the CLI.
Where do my apps and data run?
Containers, Postgres and Redis run in one region in India, on a private network with no public database endpoint. Static sites are served from Cloudflare's edge worldwide, and every request passes through Cloudflare's WAF and DDoS protection.
What happens when I hit a limit?
On Free and Hobby, deploys over a limit stop with a clear message and an upgrade link; nothing running is deleted. Pro and Team bill overage instead, with a $50 default spend cap on Pro that you can change.
Is there a trial of the paid plans?
Every new account starts on Pro for 14 days, then drops to Free unless you upgrade. You get an email 3 days before, with a list of exactly what changes.
Can I get my data out?
Yes, any time. The db tool and rivtor db export stream a standard pg_dump of your database. Your code stays in your repo; Rivtor never needs write access to it.
[
Sign in with GitHub, connect your agent, ship your app. No card, live in about 2 minutes.
Deploy