The Postgres tools are the reason to install it: pg_outliers ranks queries by resource use, pg_locks shows what is blocking what, pg_ps shows what is running, and pg_kill ends it — the same sequence you would run by hand, without the plugin invocations in between. deploy_one_off_dyno is the sharpest tool in the set: it runs your command against the real app with network and filesystem access, which is exactly as useful and as consequential as it sounds.
Heroku's own server: 33 tools over apps, dynos, add-ons, pipelines and Postgres, including a psql passthrough and a one-off dyno with network and filesystem access.
- App lifecycle: list, create, rename, inspect config and dynos and domains, and switch maintenance mode on and off
- Logs filtered by dyno, process type or source
- Postgres operations as first-class tools — SQL execution, database status, active queries, lock analysis, resource-heavy query ranking, credentials, process kill, maintenance windows, backups and version upgrades
- Dyno management: list the formation, scale quantity and size, restart a dyno, a process type or all of them
- Add-ons created and inspected, with the service catalogue and per-service plans including pricing
- Pipelines created, listed, described, and apps promoted between stages
- Deployment from an app.json, and a one-off dyno that runs a command with a size and a time to live
A Heroku account key in HEROKU_API_KEY. deploy_to_heroku additionally needs a valid app.json passed through its appJson parameter.
One command plus a key — npx -y @heroku/mcp-server, then supply credentials