Deploy your MCP server. We handle the hosting.

Bring your code or container. Deploy, manage releases and offer paid access from one place.

Deploy your MCP See how it works

Public sources supported. Hosting access and limits are shown in your publisher account.

GitHub npm PyPI Container

Notes MCP GitHub · example/notes-mcp

SourcePinned BuildBuilt ReleaseNot released

Ready for release

Gateway endpoint format https://mcprush.com/gw/<listing-id>/mcp

Illustrative preview

Add your source

Public repo, package or container image.

Build and configure

Set the runtime, transport and credentials.

Submit and release

Submit for review and manage your releases.

Run it. Update it. Keep control.

Manage your hosted runtime and releases from your publisher account.

A separate runtime for each installation

Each customer installation runs with its own configured credentials. Idle instances stop automatically.

Reconnecting after idle may require a retry while the runtime starts.

Build first. Release when you’re ready.

New builds leave your current release unchanged. Promote the next version or roll back to a recent release.

How a release goes out

Notes MCP One hosted listing

Installation A Running

Isolated runtime · Credentials configured

Installation B Idle

Stopped after idle · Credentials configured

Offer free access or paid subscriptions.

Create your listing, set subscription plans and let MCPRush handle customer access and billing.

Your subscription plans

Monthly plans and call allowances you define for your own listing.

Payments through Stripe Connect

Customer billing and publisher payouts run through Stripe.

View publisher fees and payout terms

Managed hosting included

12%

Platform share on paid sales

Account limits and Stripe processing fees apply.

Free listings can use hosting within their account limits.

Before you deploy

Public source

Use a public repository, package or supported image.

On-demand runtime

Instances stop after idle. Always-on is not available.

External storage

Use your own database or storage for persistent data.

Runtime and build details

Starter size: 0.25 vCPU, 0.5 GB. Larger sizes are opened per account.
Image: Up to 1 GB. A larger image is not accepted.
Source: Up to 250 MB unpacked.
Idle: Instances stop after about 15 minutes without calls. This is the platform default, not an account setting.
Build methods: Your own Dockerfile, Node.js, Python, or a published container image.
stdio servers: Supported on the Node.js, Python and container-image paths, where MCPRush adds the HTTP adapter. With an image you name the command that starts it. A custom Dockerfile must expose its own Streamable HTTP endpoint.
Logs: A failed build sends you the output of the command that failed, by email and in your publisher account. Runtime logs from a running server are not available to publishers yet.
Addresses and regions: Every hosted server answers at the MCPRush gateway address. Custom domains are not available, and hosted servers run in a single region.

Already have hosting? Connect your existing endpoint to MCPRush.

Hosting questions

Read the deployment guide

Do I need my own AWS account?

No. With managed hosting, MCPRush runs the infrastructure. You configure your server in your publisher account.

Can I deploy an MCP that only supports stdio?

Yes, through the supported Node.js, Python and container-image paths. MCPRush adds the HTTP adapter. With a container image you name the command that starts it, and the image needs an Alpine- or Debian-based layer the adapter can run on. Custom Dockerfile deployments must expose their own Streamable HTTP endpoint.

What happens when the server has been idle?

The runtime starts again when a customer connects. The initial request can return a starting response and require a retry. Startup time varies.

Can I use a private repository or persistent disk?

Managed builds currently require public sources: a private repository’s manifest can be read, but a managed build cannot run from one yet. Persistent local storage is not supported — use an external database or storage service for data that must survive restarts. Connecting an endpoint you run yourself keeps your code where it is.

How do updates work?

Build a new version, then choose when to release it. Building does not replace the current release, and recent previous releases stay available to roll back to.

Which MCP clients can connect?

Use a supported client through the MCPRush CLI, or configure a compatible Streamable HTTP client manually. Connection options and any required credentials are shown when the server is installed. See supported clients.

Ready to deploy your MCP?

Start with your source. Manage everything from your publisher account.

Deploy your MCP