ray.run: build and deploy any MCP server
Log in

Take any MCP server from idea to live.

Describe the server to Codex, Claude Code, or another coding agent. Rayrun builds it, verifies it, and runs it behind your endpoint.

codex · hacker-news-digest

you → Codex
Build a Hacker News digest MCP.
Add get_top_stories({ limit }), test it,
and deploy it to Rayrun.

Codex  server implemented · tests passed
$ rayrun deploy --wait
Rayrun Validating 12 source files…
Rayrun Uploading deployment source…
Rayrun Tracking build build_7f3d.
Resume  rayrun deployments get build_7f3d
Rayrun build succeeded · release active · get_top_stories

you → Claude
Use my Hacker News server to summarize
today's top five stories.

Your agent builds the whole server.

  1. rayrun init creates a working TypeScript server with ViteMCP, with a test. Add --template fastmcp for Python.
  2. Give the directory to your coding agent. Ask it to implement the whole service, not one tool.
  3. rayrun deploy --wait streams validation, upload, build, and release. It prints a resume command first, so a closed terminal loses nothing.

Any framework works. The project serves MCP at /mcp and listens on the port Rayrun supplies.

Four ways to deploy.

All four get the same tool catalogue, credentials, access policy, and Activity trail. Rayrun verifies every server before marking it ready.

ray.run/dashboard · deploy

Install the Rayrun GitHub App once. Pushes to the branch you pick build, verify, and release. Pull requests get an isolated preview.

Repositoryrayrun/hacker-news-mcp
Branchmain
Deploy a preview for every pull request

Releases fail safely.

Every revision, build, and release keeps its identity and an append-only history. Limits and rollback reference.

  • Encrypted inputs and logs

    Source and build logs are encrypted with the workspace key. Secrets never enter source, images, logs, API responses, or CLI state.

  • Verified before release

    Builds use fixed Node 24 or Python 3.13 images. The runtime is non-root. A candidate must answer tools/list before it takes traffic.

  • Failure-safe rollback

    A failed candidate leaves the current release live. Roll back to any verified build from the last 30 days.

Storage and private networks, when you need them.

  • Persistent storage

    Attach a volume to an npm package or container image. Off by default.

    • Mounted at /data. RAYRUN_DATA_DIR points to it.
    • Files survive idle stops and container replacement.
    • Up to 10 volumes per workspace. A deleted service keeps its volume for 30 days.

  • Private networks

    A managed npm server can join your Tailscale network with a tagged auth key.

    • Untagged identities are refused. Your tailnet ACLs decide what it can reach.
    • Outbound only. Nothing on the tailnet can connect in.
    • Its machine identity survives idle stops and replacement.

Changing storage or network settings replaces the container. Requests fail during that redeploy. Source-deployed services are stateless for now.

Deploy your first server.

Scaffold a project, hand it to your agent, and ship it behind your endpoint.

Built something your customers need? Publish it to them.