Manh
VISign in
The runtime for coding agents

Your coding agent builds, checks and ships web apps and websites.

One MCP connection. Database, cron, email, AI, deploys, snapshots and rollback built in. No AI of our own, no token billing, no per-end-user pricing.

Run in your terminal, then type /mcp to sign in
claude mcp add --transport http manh https://mcp.manh.run
Purchase approval: staff file requests, managers approve or reject, a dashboard by category. An agent built it from a spec; it runs on manh.run today.

Three steps from idea to a live URL

1

Connect

Add manh.run to Claude Code with one command. Sign in once; your agent only sees your projects.

2

The agent writes, manh.run checks

Your agent writes app/ on a locked stack: antd-pro for the UI, Hono + Drizzle for the API. Every write runs check, which blocks errors before users see them.

3

Deploy in one call

UI and API ship together as one Cloudflare Worker at your-app.manh.run, with its own SQLite database and a snapshot before every deploy.

Real apps, not demos

Purchase approval: staff file requests, managers approve or reject, a dashboard by category. An agent built it from a spec; it runs on manh.run today.

Try it: manager@eval.test / Eval#2026
  1. Staff file a request: title, amount, category, needed-by date.
  2. Managers get an inbox sorted by deadline, approve in bulk or reject with a reason.
  3. Every change is logged; the dashboard rolls up by category and month.
  4. The spec (PRD, ORCA, screens) sits next to the code in spec/.

What your agent no longer rebuilds

A database per app

SQLite in a Durable Object, foreign keys, migrations from the Drizzle schema.

Shared cron

A .cron.ts file gets a schedule, retries and 30 days of run history.

Email · AI · Files

Import from app/std/* — no API keys; the keys stay with manh.run.

Snapshots + rollback

The DB is captured before deploys and destructive queries; restore in one click.

Feedback from users

Users pin feedback on the exact spot in the app; your agent reads it and works through ORCA.

Checks before deploy

Framework rules block code patterns that crash apps, with a fix the agent can apply.

Not another app builder

No token billing

You already pay for your agent. manh.run has no AI of its own to resell.

No per-end-user pricing

A website with 50 or 50,000 visitors, an app with 5 or 500 users — same price.

Your code

A readable app/ tree you can edit and export. Mainstream stack, no proprietary DSL.

Questions

Which agents work?

Claude Code (recommended), the Claude app and any MCP client with OAuth. manh.run runs no LLM of its own.

Can I change the stack?

No. antd + pro-components on the front, Hono + Drizzle on Cloudflare Workers behind. The locked stack is why agents get it right fast.

Where does my data live?

Each app has its own SQLite database in a Cloudflare Durable Object, with 30 days of snapshots.

Can I take my code out?

Yes. The whole app/ and spec/ tree is plain files; a zip export button is coming.

Connect your agent in a minute.

claude mcp add --transport http manh https://mcp.manh.run
Start for free