- —Granular CPU + RAM + storage scaling per container
- —Independent dev / staging / production environment sizing
- —Shared vCPUs or Guaranteed CPUs — explicit selection
The DevOps Tax Ends Here
Legacy PaaS forced teams into oversized resource bundles. Upsun Flex lets DevOps teams provision CPU, RAM, and storage per container — independently, transparently, at scale. AI worker containers and web application containers no longer drag each other into the next pricing tier.
Four containers side by side, each labeled with a different CPU/RAM config. "LEGACY TIER" on the far left overflows its box. Three right-side containers precisely sized. Amber price tags on each.
- —Streamable HTTP transport protocol
- —"enable-write": "true" required for mutations
- —Clients: Claude Desktop · Cursor · Windsurf · Claude Code · JetBrains
Infrastructure on Command
At mcp.upsun.com/mcp, Upsun exposes its entire REST control plane as discoverable MCP tools. Claude Desktop, Cursor, Windsurf, and JetBrains AI can execute infrastructure operations from natural language. Default: read-only. Write operations require explicit authorization.
IDE window centered. Chat bubble: "Deploy my Next.js app." Containers activating in sequence — Build → Test → Route. Padlock labeled "READ-ONLY" at the entry point.
- —Three mandatory keys: applications · services · routes
- —Top-level merge only — nested properties overwrite entirely
- —No error thrown on silent overwrite — deployment fails downstream
Three Keys. No Deep Merge.
The entire Upsun deployment lifecycle is declared in .upsun/config.yaml. Three mandatory top-level keys: applications, services, routes. The parser merges files at the top level only. Nested properties in a second file overwrite the first — silently, without error. This is the most consequential thing an AI agent can misunderstand.
Filing cabinet. Three labeled drawers: APPLICATIONS / SERVICES / ROUTES. A second folder "file2.yaml" slides into APPLICATIONS and a sticky note "source.root" vanishes with a poof. Developer figure looks confused. No error message.
- —Claude skills: PR diff + Jira + Slack → Upsun-formatted Markdown
- —MCP pipeline: IDE-native doc preview before PR submission
- —Human gate: final accuracy, voice, structure, developer empathy
The Writer Who Builds the System
The Upsun content engineering mandate: don't write the docs, architect the pipeline that produces them. Claude skills ingest PR diffs, Jira tickets, and Slack context, outputting Upsun-formatted Markdown. MCP preview pipelines let engineers validate AI drafts in their IDE. One human at the final quality gate.
Rube Goldberg factory floor. Entry: PR DIFF. Conveyor belt: CLAUDE SKILL → MCP PREVIEW → ENGINEER REVIEW. Final station: one human editor with an APPROVED stamp. Amber quality gate barrier.
- —docs.upsun.com/llms.txt — structured documentation index
- —Context7 MCP integration for IDE-native doc retrieval
- —Competitor benchmark: Vercel llms-full.txt · Netlify AI context files
244 Pages. Zero Ambiguity.
Upsun rebuilt developer.upsun.com into 244+ agent-ready pages with structured frontmatter, machine-readable metadata, and a complete llms.txt index. Context7 MCP integration lets AI coding assistants retrieve docs directly from the IDE. The goal: an AI agent reading Upsun docs produces working infrastructure, not a confident hallucination.
244 book spines on shelves, each labeled with a documentation page slug. A robot arm selects exactly the right book. A hallucination ghost labeled "BAD YAML" dissolves on contact with the structured shelf. Context7 badge glows in the corner.
Every PaaS platform will add an MCP server. Every cloud docs team will adopt llms.txt. The question isn't whether agents will consume infrastructure docs — it's who builds the content architecture precise enough that an agent acting on it ships working infrastructure.
Read the Dispatch →