Use Staticbot from Claude Code or Codex — no MCP server required
One open Agent Skill gives your coding agent the API workflow, safety rules, and command-line helper it needs to deploy, migrate, and sync through Staticbot.
Claude Code and Codex already know how to use a terminal. The new Staticbot Skill turns that general shell access into a reliable Staticbot workflow: it discovers the live API schema, keeps the API key out of arguments and logs, and stops at the human approval gates that matter.
What the Skill adds beyond “just use curl”
A capable agent could construct REST calls from scratch. The risky part is not the HTTP request — it is remembering the workflow around it. A migration must pause after discovery. A destructive sync needs review. A rollback must target an exact pinned version. DNS records must be shown without casually suggesting a nameserver change that could break mail.
Live API contract
The helper fetches Staticbot’s current OpenAPI document before unfamiliar mutations, so the agent does not guess request fields.
Credential-safe requests
The bearer token stays in STATICBOT_API_KEY, absolute request URLs are refused, and redirects are not followed.
Self-navigating migrations
The agent follows pendingAction and failureBanner instead of hard-coding a brittle phase sequence.
Human decision gates
Discovery, destructive SQL, retry/skip, rollback, switchover, and deployment choices remain explicit decisions.
Install the Staticbot Skill
The source lives in the public staticbot-mcp repository. Despite the repository name, the Skill does not start or require the MCP server; the repository is the shared home for Staticbot’s agent integrations.
Codex
$skill-installer install the Staticbot skill from https://github.com/bitfiction/staticbot-mcp/tree/main/skills/staticbot
Claude Code
Install the Staticbot skill from https://github.com/bitfiction/staticbot-mcp/tree/main/skills/staticbot as a personal skill.
The Skill follows the open Agent Skills directory format: a focused SKILL.md, a deterministic shell helper, and a workflow reference loaded only when needed.
Give the agent an API key without giving it away
Open API in the Staticbot menu, or go directly to app.staticbot.dev/developer, then create an organization-scoped key in the API Keys section. Make it available to the agent process as STATICBOT_API_KEY. Do not paste it into the conversation, put it in a command argument, or commit it to the repository.
The included helper sends the key only to the configured Staticbot origin. Hosted users need no URL configuration; local and self-hosted users can set STATICBOT_API_URL.
Try these prompts
“Use $staticbot to list my deployments and explain anything waiting for DNS.”
“Use $staticbot to migrate this Firebase project to Supabase. Stop after discovery and show me exactly what will move.”
“Use $staticbot to preview this Base44 migration before switching the backend.”
“Use $staticbot to inspect the latest sync run and explain every failed job. Do not retry or skip anything yet.”
“Use $staticbot to list safe rollback targets for my website and show the commit details.”
Does this replace the Staticbot MCP server?
No. It gives shell-capable agents a lighter option. MCP remains useful when you want 50 named, typed tools exposed directly in Claude Code, Cursor, Windsurf, or another MCP client. The Skill is simpler when Claude Code or Codex already has a terminal and can inspect the live API contract itself.
| Staticbot Skill | Staticbot MCP |
|---|---|
| Claude Code or Codex with shell access | Any compatible MCP client |
| Live OpenAPI plus command-line helper | 50 typed tools and input schemas |
| No background server | Local Node.js MCP process |
See the combined Staticbot Skill and MCP documentation for both setup paths and the complete MCP tool catalog.
Deploy from the agent already in your repository
Install the Skill, expose one Staticbot API key, and ask Claude Code or Codex to handle the workflow.