Connect Claude, Cursor and VS Code to your cloud with MCP

Use the Plexivo MCP server to deploy to the cloud from Claude Code, Cursor or VS Code: create a key, add the config, and approve every plan.

7 min read

Connect Claude, Cursor and VS Code to your cloud with MCP

You already ask your editor's assistant to write the code. With the Model Context Protocol, you can ask it to ship that code too: create the app, add a database, check what is running, and tell you why a deploy failed, without leaving the editor.

Plexivo runs an MCP server for exactly this. Your assistant signs in with an API key, sees Plexivo's tools, and does only what that key allows. Anything that creates a resource goes through a plan first, and deleting stays with people in the console.

This guide covers the server details, the exact configuration for Claude Code, Cursor and VS Code, the tools your assistant gets, and the safety rules that keep your organisation in your hands.

What is an MCP server for cloud deploys?

MCP is an open protocol that lets an AI assistant call tools on a service. A cloud MCP server turns your platform's API into tools the assistant can understand: "create a database", "get deploy status", "estimate a price".

Plexivo's server:

SettingValue
URLhttps://api.plexivo.io/mcp
TransportStreamable HTTP
Protocol version2025-06-18
Sign-inAn API key in the Authorization header, as Bearer followed by the key

It works with Claude Code, Cursor, VS Code and any client that speaks Streamable HTTP. The reference is on the Connect AI tools over MCP page.

What can my assistant do on Plexivo?

When it connects, your assistant receives Plexivo's tools, each marked with its risk class.

Build and deploy

  • deploy_from_git creates a Plexivo App Runner app from a Git repository and branch.
  • create_app creates a Plexivo App Runner app from a container image or a repository.
  • create_database creates a Plexivo Postgres database, or Plexivo Database for MySQL when you ask for it.
  • create_stack creates an app and a database in one call.
  • scale_app changes an app's size and number of instances.
  • set_env replaces an app's environment variables. Connection variables Plexivo manages cannot be overwritten, and the answer lists names, never values.
  • rotate_credentials gives a database a new password. Connected apps receive it; the password itself is never returned.

Inspect and diagnose

  • get_project_context summarises your organisation: up to 10 resources with their status, and your wallet balance.
  • get_deploy_status reports on one app or database.
  • get_metrics counts your resources and how many are running.

Plan and cost

  • plan: true on any tool that changes something returns a plan instead of acting: the step, its risk class and whether it can be undone.
  • can_i says whether an action is allowed, needs a confirmation or is refused, and why.
  • estimate_architecture gives a rough monthly price in US dollars.

Account

  • list_tenants names the organisation your key belongs to.
  • get_usage_and_bill shows your wallet balance, with the Billing permission.

Deleting a resource, reading a database's connection details and running SQL happen in the Plexivo console, on the resource's page, the database's Connect page and the SQL editor.

How to connect your editor, step by step

Create an API key

Owners and admins can create keys.

  1. Go to Settings → Developer and choose Create API key.
  2. In Key name, name it after the tool and the machine, such as "Cursor on my laptop".
  3. Choose permissions. Keep Read and Create and change, the default, to let your assistant deploy. For an assistant that only looks, keep Read alone. Add Billing to let it see your wallet balance.
  4. Choose when it Expires, and set a Monthly spend limit (USD, optional) if you want one.
  5. Choose Issue key and copy it. It is shown once.

On the same page, Editor and AI tools has a ready configuration behind Copy configuration. More on permissions and limits in API keys.

Keep the key out of files you commit. Set it in your shell:

export PLEXIVO_API_KEY="paste-your-key-here"

Claude Code

claude mcp add --transport http plexivo https://api.plexivo.io/mcp \
  --header "Authorization: Bearer $PLEXIVO_API_KEY"

Add --scope user to use Plexivo in every project. Run claude mcp list to see it connected.

Cursor

In your project, create .cursor/mcp.json:

json
{
  "mcpServers": {
    "plexivo": {
      "url": "https://api.plexivo.io/mcp",
      "headers": { "Authorization": "Bearer ${env:PLEXIVO_API_KEY}" }
    }
  }
}

Cursor reads PLEXIVO_API_KEY from the environment it was started in, so start Cursor from the shell where you exported it.

VS Code

In your project, create .vscode/mcp.json. VS Code asks for the key the first time and keeps it out of the file:

json
{
  "inputs": [
    { "type": "promptString", "id": "plexivo-api-key", "description": "Plexivo API key", "password": true }
  ],
  "servers": {
    "plexivo": {
      "type": "http",
      "url": "https://api.plexivo.io/mcp",
      "headers": { "Authorization": "Bearer ${input:plexivo-api-key}" }
    }
  }
}

Any other client

Give it the URL and the header Authorization: Bearer followed by your key. Plexivo reads the key from this header only.

Check the server

Without a key, the server describes itself:

curl https://api.plexivo.io/mcp

It answers with its name, version, transport and protocol version:

json
{"name":"plexivo","version":"0.1.0","transport":"streamable-http","protocolVersion":"2025-06-18"}

How does Plexivo keep MCP changes safe?

Giving an assistant access to production deserves care. These rules apply to every call:

  • The key's permissions, nothing more. Every call passes the same sign-in, permission checks, limits and audit as the Plexivo API. A key works only in its own organisation and never does more than the person who created it can do.
  • Plans before creation. Creating resources and new passwords need a confirmation. Your assistant asks for a plan first. The plan carries a single-use confirmation token that runs only that exact call, from the same key, within 5 minutes. Ask your assistant to show you the plan before it confirms.
  • Small changes, your choice. scale_app and set_env run when asked. To stop them over MCP, an owner or admin chooses Ask me before any change or Only watch and report in Plexivo AI settings.
  • Deleting stays with people. The destroy tool always refuses and points to the Plexivo console. No setting changes that.
  • The same prices and checks. A resource your assistant creates passes the same plan, verification and wallet checks as the console, counts towards the key's monthly spend limit, and is billed at its normal price.
  • Secrets stay out of answers. Plexivo removes secret-shaped values from every tool's answer before it is sent.
  • Every change is recorded. Each change made over MCP appears in Settings → Audit log with the API key that made it. See the audit log docs.

What are the MCP limits?

MCP calls share the key's limits with the rest of the API. By default:

  • 120 requests a minute per key.
  • 4 calls running at the same time per key.
  • 20 calls a day that create resources. A plan does not count.
  • The key's monthly spend limit, if you set one.

What do MCP error codes mean?

You seeWhat to do
401The key is missing, mistyped, revoked or expired. Check the header reads Authorization: Bearer and your key.
key_scope_missingThe key lacks the permission that tool needs. Create a key with it.
confirmation_requiredNormal for creating resources. Your assistant plans first, then confirms.
autonomy_insufficientYour Plexivo AI settings stop changes over MCP. Make the change in the console, or change the setting.
human_onlyA person does this in the Plexivo console.
429The key is over one of its limits. Wait and try again.

First prompts to try

Start with questions that read, then move to plans:

  • "Which Plexivo organisation am I connected to, and what is running?"
  • "Plan a Plexivo Postgres database called orders-db." A plan creates nothing; your assistant shows you what it would do.
  • "Can I scale my api app to two instances? Use can_i first."
  • "Estimate a monthly price for this repository on Plexivo App Runner with two instances."
  • "Deploy this repository from the main branch to Plexivo App Runner. Show me the plan before you confirm."

Your app should listen on the PORT variable, 8080 by default. If you would rather talk to Plexivo in the console, by voice or text, read Ask Plexivo AI: deploy in plain language. For the terminal, see the Plexivo CLI.

FAQ

Which AI tools work with the Plexivo MCP server?

Claude Code, Cursor and VS Code have ready configurations, and any client that supports Streamable HTTP works. Give it https://api.plexivo.io/mcp and an Authorization: Bearer header with your API key.

Can my assistant delete my database?

No. The destroy tool always refuses and points to the Plexivo console, and no setting changes that. Deleting a resource is always done by a person in the console.

Will my assistant create resources without asking?

Creating a resource needs a plan and a single-use confirmation token, valid for that exact call within 5 minutes. Ask your assistant to show you every plan before it confirms. An owner or admin can also choose Only watch and report to stop changes over MCP.

How do I limit what an AI tool can spend?

Give its API key a Monthly spend limit when you create it. Plexivo checks the limit before the key creates a resource and refuses the request if it would be passed.

Is MCP usage charged?

Resources your assistant creates are billed at their normal prices, from your wallet. Each change appears in your audit log with the key that made it. Prices for every size are on the pricing page.

Ship from your editor

Create a key, paste the configuration, and ask your assistant what is running. Start your free trial on the Developer plan, with no credit card required, and connect your editor in Settings → Developer.