Deploy straight from Claude Code

SimDeploy lets Claude Code deploy a project with one non-interactive command — it analyzes the code, builds locally, publishes to the lowest-cost compatible architecture, and returns a public URL or a typed error the agent can act on.

SimDeploy was designed for agents: non-interactive CLI, JSON output, structured logs and a native MCP server. Claude Code ships a project with zero infrastructure decisions.

How it works

1. export SIMDEPLOY_TOKEN=sd_live_...  (once, by a human — never pasted in chat)
2. simdeploy analyze --json  (offline analysis: framework, architecture, estimated cost)
3. simdeploy deploy --yes  (builds locally, publishes and returns the URL)
4. simdeploy logs  (diagnosis when something fails)

—MCP server with 8 tools (analyze_project, deploy_project, get_logs, ...)
—Non-zero exit code on failure — the agent knows when to fix
—Estimated cost before publishing, labeled as a projection

How Claude Code ships with SimDeploy

Claude Code drives SimDeploy through its native MCP server and the non-interactive CLI. The agent calls analyze_project (or simdeploy analyze --json) to read the framework, routes and a cost projection offline, then deploy_project (or simdeploy deploy --yes) to build locally and publish. A successful deploy returns a public URL; a failed one returns a typed state — ANALYSIS_FAILED, BUILD_FAILED, DEPLOY_FAILED — with per-stage logs the agent reads through get_logs. Because every command exits non-zero on failure, Claude Code knows when to fix and retry without a human watching the run.

What deploys today

Static builds deploy end-to-end today, including static exports of frameworks like Next.js. Server-rendered and serverless targets are planned; asking for one returns a typed error rather than a silent partial deploy, so the agent gets a clear signal. Analysis runs entirely on your machine — only the built output is uploaded — and the scoped, revocable token stays in the environment, so Claude Code never needs the secret pasted into the conversation.

Deploy your first project Agent docs

FAQ

How does Claude Code deploy with SimDeploy?

SimDeploy ships a native MCP server and a non-interactive CLI. Claude Code calls simdeploy analyze --json, then simdeploy deploy --yes, and reads the returned public URL — or the typed error and per-stage logs if a deploy fails.

Does my code leave my machine during analysis?

No. simdeploy analyze runs offline — framework, routes and the cost estimate are computed locally. Only the built output is uploaded when you run deploy.

What can SimDeploy deploy today?

Static builds ship today, including static exports of frameworks like Next.js. Server-rendered and serverless targets are planned.

Do I have to hand Claude Code my token?

No. A human authenticates once with simdeploy login; the scoped, revocable token stays in the environment and is never pasted into the chat.