@heroku/mcp-server

Exposes Heroku Platform API operations through the Model Context Protocol, enabling LLM agents and Claude to create, read, update, and delete Heroku applications without direct API knowledge. Implements MCP resource and tool handlers that translate natural language requests into authenticated Heroku API calls, with built-in error handling and response normalization for LLM consumption.

Provides MCP tool handlers for querying, scaling, and configuring Heroku dynos (application containers). Translates dyno operations (list, describe, scale, restart) into Heroku API calls with response normalization, enabling LLM agents to manage application compute resources and monitor dyno status without direct API knowledge.

Exposes Heroku config variable (environment variable) operations through MCP tool handlers, enabling LLM agents to read, set, and delete app configuration without direct API access. Implements secure parameter passing and response filtering to prevent accidental credential exposure in LLM context windows.

Provides MCP tool handlers for triggering builds, querying build status, and managing releases on Heroku. Integrates with Heroku's build system to enable LLM agents to orchestrate deployment pipelines, monitor build progress, and rollback releases without manual intervention or direct API knowledge.

Exposes Heroku add-on operations (database, cache, monitoring services) through MCP tool handlers, enabling LLM agents to provision, configure, and deprovision add-ons without direct API access. Implements add-on discovery, plan selection, and credential extraction for seamless integration with application configuration.

Implements MCP resource handlers that expose Heroku application metadata (name, owner, region, stack, buildpacks) as queryable resources. Enables LLM agents to introspect application configuration and state without tool calls, supporting efficient context building and decision-making in multi-step workflows.

Implements standardized error handling and operation status responses across all MCP tools, translating Heroku API errors into human-readable messages for LLM consumption. Provides operation tracking for asynchronous tasks (builds, releases, add-on provisioning) with status polling support, enabling agents to monitor long-running operations without blocking.

Enables LLM agents to compose MCP tools for batch operations across multiple Heroku apps (e.g., scale all web dynos, update config across apps, provision add-ons to multiple targets). Implements app filtering and iteration patterns that allow Claude to reason about batch operations at a high level while MCP handles individual app targeting.