Grain for Cursor

# Download the desktop app from rungrain.com, then:
grain login
cd path/to/project
mkdir -p .cursor && [ -f .cursor/mcp.json ] || echo '{}' > .cursor/mcp.json
grain init
# Ask Cursor to build the artifact in this folder.
grain push
grain checkpoint -m "first Cursor artifact"
grain share --public
# https://<slug>.share.rungrain.com

What the agent does, what Grain does

Setup

1. Download the desktop app from rungrain.com, open it and sign in. The download starts directly, so you can install Grain and connect your existing agent right away.
2. In a terminal, run grain login. Change into the project folder that Cursor opens. Connect the narrowest folder containing the artifact rather than a broad directory with unrelated material.
3. Create .cursor/mcp.json containing {} if the project does not have one, then run grain init. It adds the Grain server to .cursor/mcp.json and keeps other servers. Without that file, init writes only .mcp.json, which Cursor does not read. Open .cursor/mcp.json if the server is missing in Cursor.
4. Open the folder in Cursor and confirm the connection in the client. Give agent mode a prompt that names the reader, deliverable and completion checks. Say whether the result is static or needs Postgres, login or a scheduled job.
5. Keep Cursor as the editor. Grain is downstream from the editing loop, so you can inspect diffs and run project checks before the files are pushed to the artifact workspace.

First artifact

1. Start with a bounded prompt: Build a working customer prototype for the onboarding flow in this folder. Include a three-step form, a confirmation view and a short assumptions panel. Use sample data, require no login, and make every limitation visible on the page. This gives Cursor a product task instead of the vague request to make a demo.
2. Inspect the files and run the repository's checks. Ask Cursor to remove credentials, local file paths, copied logs and claims that are not supported by the source material. Decide whether sample data is enough before introducing Postgres.
3. Run grain push, followed by grain checkpoint -m "first Cursor artifact". The checkpoint records the published state after review. It does not replace the repository history you use for the source code.
4. Run grain share --public when anyone with the link may view it. Use grain share --password <pw> for a password boundary, or invited-person sharing when access belongs to specific people.
5. Open the returned https://<slug>.share.rungrain.com address in a signed-out browser. Complete the main path as the customer would and confirm that the chosen sharing boundary behaves as expected.
6. Send the live address and request comments on the artifact. Feed those comments back into Cursor, revise the same folder, then push and checkpoint again. The artifact stays at a useful address while its files change.

What you pay

Your Cursor subscription and model choices remain yours. Grain does not bundle a coding agent or add a token charge. The free tier has no card requirement and includes unlimited published artifacts, ten sandboxes that sleep when idle, one always-on workspace, the desktop app, CLI and open source code under AGPL-3.0. Pro is $65 per user per month for more always-on workspaces, team seats, shared spaces and CNAME custom domains. Enterprise is offered by contact for the customer's cloud, SSO, audit and policy controls and direct support. You pay for Grain capabilities only when the free publishing boundary is not enough.

Limits and known gaps

• Grain is not an IDE replacement. It does not provide Cursor's editor, agent mode or human code-navigation workflow.
• Cursor remains responsible for creating and changing the project. Publishing a result does not prove that the implementation is correct or secure.
• The integration file is .cursor/mcp.json. A policy-managed Cursor setup may require review before that file can change.
• A public share is available to anyone with the link. Use a password or invitations for artifacts that should have a narrower audience.
• The desktop app runs on macOS, Windows and Linux. Without the desktop app, add the hosted MCP server instead. It can find, create and edit HTML artifacts and run commands in app workspaces, but it cannot create share links.
• Use a general deploy platform when the job is a production service with its own repository pipeline. Grain publishes agent output directly and is not a general deployment platform.

Frequently asked questions