Heroku MCP

The Heroku MCP server puts the Heroku Platform behind a set of named tools an AI client can call. App tools cover list_apps, get_app_info, create_app, rename_app and transfer_app, plus deploy_to_heroku for app.json driven deploys and deploy_one_off_dyno for running code in a sandboxed dyno. Process tools include ps_list, ps_scale and ps_restart for sizing and restarting dynos. Add-on tools list, describe and provision add-ons. Operational tools cover maintenance_on, maintenance_off and get_app_logs, while pipeline tools create pipelines, list them, read their details and promote apps to the next stage. Team and space tools enumerate teams and private spaces. Postgres gets its own group: pg_psql runs SQL, pg_info reports database state, pg_ps shows active queries, pg_locks exposes lock contention, pg_outliers surfaces the most resource intensive queries, and pg_credentials, pg_kill, pg_maintenance, pg_backups and pg_upgrade handle credentials, process termination, maintenance windows, backup schedules and version upgrades. Setup has two paths. Starting the server through the Heroku CLI reuses the credentials already on the machine, so no API key is pasted into a config file; alternatively the npm package runs with a Heroku authorization token in the environment. The server keeps the CLI in REPL mode so calls do not pay process startup cost each time. It fits developers using AI coding clients across Common Runtime, Cedar, Shield and Fir private spaces.