/manufact-cloud

name: manufact-cloud
description: Deploy and manage MCP servers on Manufact Cloud with the mcp-use CLI, including authentication, projects, environments, deployments, domains, organizations, publishing checks, and YAML alerts.

Manufact Cloud CLI

Use the CLI executable selected by the user (normally `mcp-use`). Cloud commands work with any MCP framework. Check `mcp-use <command> --help` before using unfamiliar flags; this skill ships with the CLI version whose commands it describes.

Install or refresh the bundled guidance with `mcp-use skills install`. It copies the installed version's skills into `.agents/skills`, `.claude/skills`, and `.cursor/skills` in the selected project and works offline.

Choose output for the task

• For a human reviewing results or taking screenshots, omit `--json`. Commands

print readable summaries and tables; narrow tables become labeled records. Deploy and checklist waits show a spinner in an interactive terminal.

• For an agent parsing results or a CI script, use `--json` and the exit code.

JSON output stays machine-readable and suppresses spinners. Do not parse terminal columns or depend on colors or human wording.

• Checklist reports group results by category, expand findings and fix guidance,

and show **Client Readiness**. `checklist get <run-id> --verbose` includes all passing checks and full finding details. Read the checklist reference below before interpreting readiness or a raw failed check as a publishing blocker.

Agent workflow

• Pass `--json` when consuming output. A command writes one JSON value to

stdout, including errors. Exit 0 means success, 1 operational failure, and 2 invalid or missing input. Errors contain `error.code`, `error.message`, and `error.nextSteps` with commands to resolve the problem.

• In CI, provide `MANUFACT_TOKEN` through the secret store. `MCP_USE_API_KEY`

is the fallback. Check identity with `mcp-use whoami --json`.

• Without a terminal, `mcp-use login --json` returns a device code and exits.

Present the verification URL and user code, then run the returned `login --device-code <code>` command after approval. An onboarding prompt may already include an approved device code.

• Resolve project context before changing a server. `link --server <id-or-slug>`

saves `.manufact/project.json`. Server selection is `--server`, `MANUFACT_SERVER`, then that link. Organization selection is `--org`, `MANUFACT_ORG`, the link, the saved active organization, then the account default. Use `org use <id-or-slug>` to change the active organization.

• `deploy` waits for completion by default. With `--no-wait`, retain the

returned deployment ID and inspect `deployments get <id> --json` later. Use `logs <id> --build` to diagnose a failed build or start.

• Use the returned `dashboardUrl` when a human needs the dashboard. Keep

credentials and environment variable values out of chat, logs, and commits.

Report CLI feedback

When the user asks you to report a CLI problem or suggestion, submit a concise report with `mcp-use feedback send '...' --json`. Pass the message as one literal argument, using an argument array or proper shell escaping so backticks and `$` in report text are not evaluated. For text beginning with a dash, use `mcp-use feedback send --json -- '...'`. Include expected and observed behavior and the affected command; omit secrets, personal data, raw logs, and conversation contents. Ask before sending when the user has not authorized a report.

No login is needed. Explicit feedback is sent even when passive usage analytics is disabled. Exit 0 means the capture endpoint accepted the request; retain the `reportId`. It does not confirm human review. Inspect `error.code` on failure: `feedback_rejected`, `feedback_network_error`, or `feedback_timeout` (exit 1). Invalid input exits 2 with `usage_error`; correct the arguments before sending. Network failures and timeouts leave acceptance unknown; do not retry automatically.

Choose a command

• **Deploy or redeploy:** read [deployment workflows](references/deployments.md).
• **Environment variables and configuration:** read

[server management](references/management.md).

• **Domains, GitHub, organizations, and publishing checks:** read

[other Cloud commands](references/cloud-commands.md).

• **Create or update alerts:** read [alert rules](references/alerts.md) for the

dashboard YAML format, metric selection, and useful recovery conditions.

• **Analyze server usage, errors, latency, or clients:** use the bundled

`manufact-analytics` skill for `analytics catalog` and read-only SQL queries.

`rm` is the destructive verb, with `remove` as an alias. Confirm the target organization and server from the user's request before running destructive commands. Use existing authorization when the user already requested the change.