Heroku MCP Server

The Heroku Platform MCP Server connects an AI assistant to the Heroku cloud platform. Instead of looking up Heroku CLI commands or clicking through the dashboard, you describe in plain language what should happen: "Show me the latest logs of the production app", "Scale the web dyno to two instances", or "Create a review app for this branch". The assistant translates that into calls to the matching tools and works against your real Heroku account. The project is maintained by Heroku itself, is open source, and, according to the README, is still in an early stage of development.

What is the Heroku MCP Server?

The server implements the Model Context Protocol (MCP) and runs locally on your machine as a process that the AI client starts. It uses the Heroku CLI underneath and therefore reaches the same platform functions you would otherwise call in the terminal. According to the provider, it works with Common Runtime as well as Cedar Private Spaces, Shield Spaces, and Fir Private Spaces. Documented clients are Claude Desktop, Zed, Cursor, Windsurf, Cline, VS Code, and Trae, but in principle it can be added to any MCP-capable client.

Features and tools

The tools cover everyday work on the platform. App management includes listing, creating, renaming, and transferring apps as well as deployments and one-off dynos. For processes there are listing, scaling, and restarting. Add-ons can be listed, looked up, and provisioned. On top of that come maintenance mode, fetching app logs, pipelines (create, list, details, promotion), and an overview of teams and private spaces.

The Heroku Postgres area is particularly extensive. The server can show database information, inspect running queries and locks, find slow queries ("outliers"), manage credentials and backups, trigger maintenance and upgrades, and even run SQL through pg_psql. For troubleshooting this is a real advantage, because the assistant can run several diagnostic steps in a row and put the results into context straight away.

Setup

The prerequisite is the Heroku CLI in version 10.8.1 or higher. The recommended route is the command heroku mcp:start: it reuses your existing CLI login, so no API key ends up in a configuration file. Alternatively, you start the server with npx -y @heroku/mcp-server and store an authorisation token in the environment variable HEROKU_API_KEY. You create the token with heroku authorizations:create, with heroku auth:token, or in the dashboard under account settings. Example configurations for the clients mentioned are in the repository and in the Heroku documentation.

Costs and licence

The server is licensed under Apache 2.0 and costs nothing by itself. What matters are the resources you create or scale through it: dynos, add-ons, and databases are billed according to the plans of your Heroku account. Heroku publishes current terms on its official pricing page. Keep in mind that an assistant can trigger actions such as scaling or provisioning an add-on that cause follow-on costs.

Who is the Heroku MCP Server for?

It suits developers who already deploy to Heroku and want to handle routine tasks such as log analysis, scaling, or database diagnostics from within their development environment without switching context. Small teams without a dedicated platform role benefit as well, because the assistant knows CLI syntax and Postgres diagnostic commands. If you do not use Heroku, you gain nothing, since the server is tailored exclusively to this platform. For automated production workflows, classic pipelines and infrastructure as code remain the more dependable choice.

Security and limits

The server acts with the permissions of your Heroku account or token. It can therefore change apps, scale them, switch them into maintenance mode, and work on databases. Review actions with consequences before you confirm them, and use a staging app for experiments. Prefer heroku mcp:start so that no token ends up in configuration files or prompts. The maintainer points out that the scope and the tools may still change.

Frequently asked questions

Is the Heroku MCP Server free? The server itself is, because it is open source. Costs arise from the Heroku resources you use with it.

Do I need an API key? Not necessarily. With heroku mcp:start your existing CLI login is enough, only the npx route requires HEROKU_API_KEY.

Can the assistant also modify databases? Yes, the Postgres tools extend to backups, maintenance, and SQL queries. Work carefully and with staging data.

Heroku account, Heroku CLI version 10.8.1 or higher (installed globally), Node.js for the npx route, and an MCP client (Claude Desktop, Cursor, Windsurf, Cline, VS Code, Zed, Trae, etc.).

The Heroku MCP Server lets AI clients such as Cursor or Claude Desktop manage Heroku apps, logs, and Postgres databases. This guide shows the fastest path to a first working connection.

Step 1: Check the prerequisites

Install the Heroku CLI globally and confirm with heroku --version that at least version 10.8.1 is running. Sign in with heroku login so that the CLI session is valid.

Step 2: Connect the server via CLI login

Enter the command heroku mcp:start in your client's MCP configuration, for Cursor for example under mcpServers with the name heroku. This route reuses your existing login, so no token sits in a file.

Alternative with an API token

If the CLI session is not available, for example in an isolated environment, start the server with npx -y @heroku/mcp-server. Create a token beforehand with heroku authorizations:create and pass it as the environment variable HEROKU_API_KEY, never in a prompt or repository.

npx -y @heroku/mcp-server

Step 3: Test the connection

Start with a read-only request, such as "List my Heroku apps" or "Show the latest logs of my staging app". Only once that works is it worth moving on to scaling or database diagnostics.

Set limits deliberately

The server acts with the full permissions of your account. Practise on a staging app, confirm write actions one by one, and check possible follow-on costs before scaling or adding new add-ons.

Existing Heroku CLI session with heroku mcp:start; alternatively an authorisation token in the environment variable HEROKU_API_KEY.

Permissions follow the signed-in Heroku account or the token. The tools can read and modify apps, dynos, add-ons, pipelines, and Postgres databases.

The server runs locally and calls the Heroku platform APIs through the CLI. App logs, configuration, and query results enter the AI client context and therefore reach the respective model provider.

Write actions (scaling, maintenance mode, deleting or transferring apps, SQL on production databases), accidental follow-on costs, an API token stored in plain text, and sensitive data in logs and query results.