Runtime for Claude Code
Claude Code support is coming soon as one harness inside Spinup's cloud agent runtime, not a Claude-native managed-agent box. The Spinup Agent sits above the harness and the model: isolated environment, projected secrets, persisted state, MCP reach, and cloud-shaped approvals. The harness inside is swappable. The model it talks to can change. Today, OpenClaw and Hermes already run on this model.
Coming Soon
What Claude Code needs from the runtime
Claude Code has its own configuration model, permission system, and extension layer. Each one has to be handled explicitly before Spinup can call it supported.
Anthropic API key projection
Claude Code reads its API key from the environment. The work is to project it as a scoped secret using the same workspace model that already runs OpenClaw and Hermes.
.claude state that survives runs
Claude Code stores project context, commands, and agent memory in the .claude directory. That state needs a persistence contract before runs can resume reliably.
MCP reachability
Claude Code extends its capabilities through MCP servers. Reachability and network policy need a clear shape before MCP becomes a product promise.
Unattended approval behavior
Claude Code's permission model is interactive in places. Spinup `/exec` is unattended, so the approval flow needs a cloud-shaped contract before launch claims expand.
Why It Matters
Claude Code in the cloud is a different problem than Claude Code on your laptop
On a developer's laptop, Claude Code works because the environment is already set up. The API key sits in the shell profile. The .claude directory lives in the project root. MCP servers run on localhost. File permissions match the user account. None of that transfers to the cloud automatically.
When you move Claude Code to a cloud environment for CI pipelines, background tasks, or team-shared agents, you need to reconstruct those assumptions. The API key needs secure injection. The .claude directory needs to survive process restarts. MCP servers need network routing. File and shell permissions need an outer security boundary that does not depend on a human clicking "allow" in a terminal.
You can't run Claude Code on Spinup yet. OpenClaw and Hermes are the supported harnesses today, on the same isolated environment, projected secrets, and persistent state model. Claude Code joins them once API key projection, .claude persistence, MCP reach, and unattended approvals each have a real shape.
FAQ
Claude Code runtime questions
How does Spinup handle Anthropic API key management for Claude Code?+
Not yet. Claude Code is not a supported harness on Spinup. When it lands, the Anthropic API key will be projected as a scoped secret using the same workspace model OpenClaw and Hermes use today.
Does the .claude directory persist between runs?+
Not yet. Until Claude Code is a supported harness, Spinup makes no promises about .claude persistence, CLAUDE.md projection, or snapshot restore for it.
Can Claude Code connect to MCP servers from inside the runtime?+
Not yet. MCP behavior is part of the Claude Code work that comes after launch. The OpenClaw and Hermes paths do not cover it.
How do tool approval workflows interact with Spinup's controls?+
That contract still needs to be designed. Claude Code's interactive approval flow has to translate into something unattended before it can sit behind `/exec`. Today, launch support is OpenClaw and Hermes inside isolated microVMs.
Related
How Spinup fits into the broader runtime picture
Claude Code is on the roadmap. These pages cover what's already running: the runtime model, the managed-agent framing, and the OpenClaw and Hermes paths.
Early access
Claude Code is on the way. Try OpenClaw or Hermes on the same runtime today.
Request access if this is the runtime shape your team has been missing.