redu MCP · Model Context Protocol

Deploy and operate with your agent.

Connect the redu Model Context Protocol server to the agent you already use. Deploy a project, change its configuration, or investigate a live failure in the same conversation.

More on the platform: the cloud your agent deploys to and operates.

Add to your agent
claude mcp add --transport http redu https://mcp.redu.cloud/mcp --client-id redu-mcp

Works with any MCP client, Claude Code and Codex shown here. A browser opens so you sign in to your redu account.

Get started freeView API docs
How it works

Connect, deploy, keep operating.

01

Add the redu MCP

Run the command above to add it to your agent. A browser opens so you sign in to your redu account. You get £200 credits to start.

02

Point it at a project

From your project folder: "Deploy this project to redu." Your agent reads the repo and works out the infrastructure it needs.

03

Approve the plan, get a URL

Your agent shows you the resources and cost, checked against your quota. Approve the plan, and it deploys your application. redu checks deployment health before marking it ready at a live URL.

04

Keep working with your agent

Ask it to update configuration, add capacity, or read the logs and investigate a failure. It can work on the running infrastructure, with you steering the changes.

Example workflow

Deploy, then keep working.

Approve a deployment, then update the running app from the same conversation.

Claude Code
You
Deploy this project to redu.
Agent
Scanned the repo: a Node app on port 3000, needs Postgres.

Plan:
  - App VM (m1.large), eu-west
  - Managed Postgres (m1.small), private, auto-wired
  - Public URL: https://your-app.redu.cloud
  - ~£0.07/hr, billed hourly

Approve to provision?
You
Approve.
Agent
✓ Provisioned the VM and the managed Postgres
✓ Built and started the container
✓ Health check passed on :3000

Live: https://your-app.redu.cloud
You
Update the app to the latest commit from my repo.
Agent
✓ Latest commit deployed
✓ Deployment health checks passed

Live: https://your-app.redu.cloud

Our public benchmark compares deployment and live operations across redu, AWS, Google Cloud and Azure: 124 published Umami runs with Claude Opus 5, with results and instructions to reproduce them.

Watch real deploys →The same deploy on four clouds →View the benchmark →
Available tools

30+ tools across 6 categories.

Deploy
  • plan_deployScan a project and return a sized, costed, quota-checked deploy plan.
  • deploy_appBuild a repo or upload on a VM and serve it at https://<name>.redu.cloud.
  • deploy_composeDeploy a docker-compose project onto a VM and expose a service.
  • check_deploy_prerequisitesAuto-select a network and keypair before a deploy.
  • prepare_uploadUpload local or uncommitted code when there is no git remote.
  • get_deploymentCheck status, read the build log, and get the deploy report.
Managed data
  • create_databaseProvision a managed Postgres on its own VM, auto-wired to your app.
  • create_redisProvision a managed Redis instance.
  • create_clickhouseProvision a managed ClickHouse instance.
  • create_relational_databaseProvision a managed MySQL or MariaDB instance.
  • list_databasesList managed databases with status and connection details.
Instances
  • list_instancesList all instances with status, IP, and uptime.
  • create_instanceLaunch a VM: size, image, keypair, DNS, and volume.
  • instance_actionStart, stop, reboot, or resize a running instance.
  • get_ssh_commandGet the SSH command with the forwarded TCP port.
  • get_instance_logsRead an instance's console and boot logs.
Storage & backups
  • create_snapshotSnapshot an instance before a deploy or migration.
  • list_volumesList block storage volumes with attachment status.
  • create_volumeCreate a volume and attach it to an instance.
  • list_backupsList volume backups.
Networking
  • list_private_networksList private networks with CIDR and status.
  • list_security_groupsList security groups and their rules.
  • list_dns_entriesList DNS proxy host entries.
  • list_domainsList verified custom domains on your account.
Account & catalog
  • whoamiConfirm your credential and see your account.
  • list_flavorsList instance sizes and hourly pricing.
  • list_clustersList autoscaling clusters with status.
  • import_keypairRegister your SSH public key before creating instances.
  • list_regionsList available EU deployment regions.
Example prompts

Things you can say to your agent.

“

Deploy this project to redu.

“

Show me the deploy plan and cost before you provision.

“

Take a snapshot before I redeploy.

“

Update the app to the latest commit from my repo.

“

Show me the cost of adding another app instance.

“

Update my app's configuration and check it still responds.

Build on it

Build your own on top of the MCP.

Every tool here is also a plain API call, so you are not limited to what we ship. Wire it into your own workflows: auto-deploy on every push, tear down preview environments nightly, spin up a fresh stack for each pull request, scale up under load and back down when it drops. That is what an agent-native cloud unlocks. Your infrastructure is programmable, by your agents or your own code.

Read the API docs →
Autocoding agents (experimental)

Or let a fleet of agents write the code.

The same connection can spin up a fleet of agents that read tasks from your repo, implement them, run your tests, and open pull requests, overnight, in parallel, using your own Claude subscription (no per-token API charges).

01

Set up the fleet

In Claude Code: "Set up autocoding agents for my repo with 3 workers." It provisions a controller plus worker VMs on your account and seeds a starter ROADMAP.

02

Add tasks to ROADMAP.md

List scoped tasks and mark each one (agent-ready). Each task becomes one focused pull request, best for endpoints, tests, bugfixes, and UI changes.

03

Review the pull requests

Workers pick up tasks in parallel, write the code, run the tests, and open PRs on GitHub. Review and merge like any other PR. After each merge, the fleet learns.

Full setup guide →
Common issues

Setup troubleshooting.

claude mcp list shows nothing even though the server is configured

Some Claude Code builds silently ignore servers stored under projects[...].mcpServers in ~/.claude.json, which is where claude mcp add writes by default. No error is emitted; the tools just never load.

Fix: add it with user scope so it is stored globally:

claude mcp add --transport http -s user redu https://mcp.redu.cloud/mcp --client-id redu-mcp

Connecting with an API key instead (for CI or a committable team config)? A .mcp.json in your project root also avoids the issue:

{
  "mcpServers": {
    "redu-cloud": {
      "type": "http",
      "url": "https://mcp.redu.cloud/mcp",
      "headers": { "Authorization": "Bearer ${REDU_API_KEY}" }
    }
  }
}
Set REDU_API_KEY as an env var so your key never lives in the file.
Should I commit the API-key .mcp.json to git?

Connecting with OAuth (the default) needs no .mcp.json and no key, so there is nothing to commit. This only applies to the API-key path.

If you do use it: committing is fine when you keep the ${REDU_API_KEY} placeholder. The file holds no secret; each teammate sets their own key as an env var and picks up the server automatically. If you hard-code the key instead, add .mcp.json to .gitignore.

curl returns Bad Request: initialize first on Windows / PowerShell

PowerShell mangles inline JSON in -d '...'. Save your request body to a file and use curl.exe (not the PowerShell alias) with --data-binary:

curl.exe --data-binary "@body.json" https://mcp.redu.cloud/mcp ...

Point your agent at redu.

Deploy your project, then use the same agent to make changes and investigate issues. Billed by the hour, £200 credits to start.

Start for free · £200 credits included