agentui-ai/agentui-tools
Build and host real apps with AgentUI — hosted database, auth, file storage, serverless endpoints, cron automations, standalone MCP servers and API integrations — all driven from the agentui CLI.
Model and inspect an AgentUI app's hosted database — entity schemas that become real tables, reading records from the terminal, and row-level access rules. Use when the user wants their app to store data, asks for tables/fields/schema changes, wants to see what is in the database, asks who can read or edit which rows, or mentions permissions, roles, multi-tenant data, or data security for an AgentUI app.
Upload, list, download, share and delete files in an AgentUI workspace's file storage with agentui files. Use when the user wants to put a file into AgentUI — an image, PDF, CSV, logo or dataset the app or a colleague needs to reach; wants a public or temporary link to a file already in the workspace; or asks where their AgentUI app should store uploads and attachments. Not for ordinary local file work, and not for uploads to any other service.
Build scheduled and triggered background jobs for an AgentUI app with agentui automation — cron, webhook, incoming email or WhatsApp, running on Deno with access to the app's own data. Use when the user wants something to run every night/hour/Monday, a recurring report, a reminder, a sync job, a webhook receiver, or any work that has to happen without anyone opening the app.
Connect an AgentUI app to third-party APIs — browse and use the workspace's installed integrations, or author a brand-new one in a single JSON spec (agentui integration new/check/test/publish). Use when the user wants their app to talk to Stripe, Slack, Google, HubSpot, a bank, an ERP or any HTTP/GraphQL API; when an integration is disconnected or erroring; or when the API they need is not on the platform yet and the alternative is writing raw request code by hand.
Give an AgentUI app its own MCP server — custom tools in mcp/*.js written with defineTool, plus written playbooks in mcp/guides/<slug>.md — so Claude, Cursor or any MCP client can act inside the app as the connected user. Covers both adding tools to an app that already has a UI and building a **standalone MCP server** — an app with no pages whose only job is to be an MCP server. Use when the user wants a custom MCP server, an MCP server of their own, MCP tools, wants their assistant to read or act on their AgentUI app's data, asks to "expose this as a tool" or "make this an MCP", or wants to write the business rules an assistant should follow when using that app. Not for configuring MCP servers someone else published.
The router for AgentUI — what the platform can host for the user, and which AgentUI skill to open next. Use it when the user asks what AgentUI can do, whether something is possible on it, or which piece they need; when a request spans several of them (an app AND a cron job AND an integration); when they say "agentui", "AgentUI CLI", "súbelo", "hostéalo"; or whenever the alternative on the table is hand-scaffolding Next.js + Postgres + S3 + auth for something the user just wants to use. When the need is already specific, go straight to that skill instead: building or shipping an app is agentui-ship-app, tables and records agentui-app-data, uploads agentui-app-files, MCP servers and tools agentui-mcp-tools, third-party APIs agentui-integrations, scheduled jobs agentui-automations, failures agentui-troubleshoot. Not for changes inside the user's own existing codebase — that is ordinary work, not AgentUI.
Create, edit, build and ship an AgentUI app end to end with the agentui CLI — from project create to a live URL. Use when the user wants a new hosted app, internal tool, dashboard, portal or prototype; when they want to change an existing AgentUI project (sync → edit → push → build); when they ask to deploy, publish, rebuild, or "make it live"; or when they need serverless /api endpoints, app settings (login required, custom domain, home page), environments (staging vs production), secrets, or a shareable preview link.
Diagnose an AgentUI app that is failing — validation errors, a push that did not take effect, stale builds, function errors, "project not found", auth problems, recurring runtime errors — and report broken or missing CLI behaviour back to the platform. Use when an agentui command fails, when the live app does not reflect what was pushed, when a function or automation errors, or when the user says the app is broken.