PlatformComparing to Ink

Ink vs Heroku

How Ink compares to Heroku — from legacy PaaS to agent-native infrastructure

Heroku pioneered platform-as-a-service. git push heroku main changed how developers deployed code. But Heroku was designed in 2007 for a world where humans typed commands and read dashboards.

Feature comparison

InkHeroku
Agent integrationSkill (prompt-guided), MCP (Streamable HTTP), CLIMCP via local Heroku CLI
MCP capabilitiesFull read/write: deploy, delete, scale, databases, DNS, logs, metricsApp lifecycle, Postgres ops, add-ons, pipelines, teams
InfrastructureBare metal (self-owned dedicated servers)AWS (cloud VMs)
Build systemRailpack auto-detection, Dockerfile, StaticBuildpacks (Cedar), Cloud Native Buildpacks (Fir)
Pricing modelPer-minute compute billing, no seat feesPer-dyno billing + database fees
DatabasesPostgreSQL, Redis, MySQL, MongoDB (via templates)Paid Postgres, Redis, and Kafka add-ons
Sleep behaviorNo sleep — services run continuouslyEco dynos sleep after 30 min inactivity
WebSocketsNative support, no platform request timeoutSupported, with router inactivity behavior
HTTP timeoutNo enforced platform request timeoutRouter request timeout behavior applies
DNS managementFull programmatic DNS via MCP (A, AAAA, CNAME, MX, TXT, CAA)No DNS hosting — external provider required
Docker supportDockerfile build packContainer Registry + heroku.yml
GraphQL APIYes, with introspectionNo

Capabilities checklist

CapabilityInkHeroku
MCP server
Agent Skill (prompt-guided)
CLI
No CLI required for MCP
Multi-agent collaboration
Deploy via MCP
Delete services via MCP
Provision databases via MCP
DNS management via MCP
Metrics via MCP
Logs via MCP
GraphQL API
Bare metal infrastructure
No HTTP request timeout
No sleep/cold start on any tier
Per-minute billing
Add-on marketplace

Where the gap is real

Agent integration: one path vs three

Heroku's MCP server wraps the Heroku CLI. It runs as a local stdio process requiring the Heroku CLI and heroku login authentication. Agents can manage apps, scale dynos, read logs, and operate Postgres — a solid set of tools. But it's the only integration path.

Ink gives agents three ways in. The Skill teaches agents the Ink CLI through a prompt file — zero infrastructure, the simplest setup for Claude Code users. The MCP server at https://mcp.deployink.com/ exposes 30+ tools over Streamable HTTP — no CLI needed, works with any MCP client. The CLI works standalone for scripts and CI/CD.

Router timeouts

Heroku's router timeout model means long-running web work usually needs streaming progress or a background worker dyno. You can stream AI responses if data flows regularly, but silent long computation and idle connections need architectural workarounds.

Ink runs persistent processes with no enforced HTTP timeout. Streaming, long-polling, SSE — all work without architectural workarounds.

Eco dyno sleep

Heroku's low-cost shared dyno path sleeps web dynos after inactivity and is limited by shared monthly hours. When hours are exhausted, dynos sleep for the rest of the month.

Ink services run continuously. No sleep behavior, no cold starts, no hour pools.

DNS is external

Heroku doesn't host DNS. You register a domain elsewhere, host DNS elsewhere, and configure CNAME records pointing to Heroku's DNS targets. TLS is automated via ACM, but the DNS layer is entirely your responsibility.

Ink manages DNS programmatically through MCP. Delegate your zone to ns1.ml.ink and ns2.ml.ink, and your agent creates A, AAAA, CNAME, MX, TXT, and CAA records directly. No external DNS provider needed.

Infrastructure cost

Heroku runs on AWS. You pay Heroku's platform margin on top of the underlying cloud cost, with dyno pricing varying by dyno class and database fees billed separately.

Ink runs on bare metal — self-owned dedicated servers. No cloud provider markup. Per-minute billing for actual compute consumed.

What Heroku does well

Heroku's add-on marketplace is unmatched. The Fir generation brings Kubernetes-based infrastructure with ARM processors. And git push heroku main remains one of the most intuitive deployment experiences ever designed — for humans.

On this page