What it does

The Heroku Platform MCP Server bridges LLMs and the Heroku Platform, providing 38 tools for reading, managing, and operating Heroku resources directly through natural language. It authenticates either through your existing Heroku CLI session (recommended via heroku mcp:start) or through API key environment variables. The server supports Common Runtime, Cedar Private and Shield Spaces, and Fir Private Spaces deployments. It remains in early development, with functionality and available tools expected to evolve.

Who it's for

Backend engineers and DevOps teams deploying applications on Heroku who want to manage, monitor, and troubleshoot their infrastructure through Claude or other MCP-compatible LLM clients without leaving the editor.

Common use cases

Query application status, logs, and metrics from Heroku directly via Claude
Deploy, scale, or restart Heroku applications through natural language commands
Configure environment variables and manage add-ons for deployed apps
Monitor application health and troubleshoot issues in context
Perform CRUD operations on Heroku resources without switching tools

Setup pitfalls

Requires Heroku CLI v10.8.1 or higher — verify with heroku --version before configuring the MCP server
When using npx @heroku/mcp-server instead of heroku mcp:start, you must generate and set HEROKU_API_KEY via heroku authorizations:create or heroku auth:token — mismatched or expired keys will cause authentication failures
The server reads and writes to your filesystem — ensure your MCP client sandbox or file permissions align with the Heroku configuration directories it accesses
Early development status means breaking changes are possible; monitor releases regularly

Tool name

Description

Destructive?

pipelines_create

✓ no

pipelines_promote

✓ no

pipelines_list

✓ no

pipelines_info

✓ no

pipelines

✓ no

list_addons		✓ no
get_addon_info		✓ no
create_addon		✓ no
list_addon_services		✓ no
list_addon_plans		✓ no
maintenance_on		✓ no
maintenance_off		✓ no
list_ai_available_models		✓ no
provision_ai_model		✓ no
make_ai_inference		✓ no
deploy_to_heroku		✓ no
deploy_one_off_dyno		✓ no
get_app_logs		✓ no
list_teams		✓ no
list_apps		✓ no
get_app_info		✓ no
create_app		✓ no
rename_app		✓ no
list_private_spaces		✓ no
ps_list		✓ no
ps_scale		✓ no
ps_restart		✓ no
pg_psql		✓ no
pg_info		✓ no
pg_ps		✓ no
pg_locks		✓ no
pg_outliers		✓ no
pg_credentials		✓ no
pg_kill⚠		⚠ yes
pg_maintenance		✓ no
pg_backups		✓ no
pg_upgrade		✓ no
test_tool		✓ no

@heroku/mcp-server

What it does

Who it's for

Common use cases

Setup pitfalls