# Rowly

> The AI-native backend for SMEs. Define business entities as docs, expose them automatically to AI agents via the Model Context Protocol. Built on Cloudflare Workers + D1.

Rowly is a backend framework where you define data schemas ("docs") once and they compile to: a database table, a Zod validation schema, permission rules, a workflow state machine, an MCP tool interface, and an admin UI — all in sync, no drift.

## Documentation

- [Overview](/docs): What Rowly is and how the docs fit together.
- [Getting started](/docs/getting-started): Sign up, create an app, define your first doc, and invoke it from Claude Desktop in five minutes.
- [Doc spec](/docs/doc-spec): Twelve field types, four naming modes, and the schema/meta split that compiles to all 8 runtime artifacts.
- [MCP tools](/docs/mcp-tools): Current MCP catalog over Streamable HTTP — core tools, Thai accounting, and extensions.
- [API keys & auth](/docs/api-keys): Bearer tokens with user- and app-scope, permission scopes, and the show-once discipline.

## Quick start for AI agents

To connect an MCP client (Claude Desktop, Cursor, etc.) to Rowly:

1. Sign up at https://rowly.ai/signup (email OTP, no password)
2. Create an app in the admin dashboard
3. Define a doc (e.g. Vehicle with fields: plate, daily_rate, status)
4. Create an app-scoped API key at /api-keys (shown once — copy it)
5. Add to your MCP client config:

```json
{
  "mcpServers": {
    "rowly": {
      "url": "https://rowly.ai/mcp",
      "headers": { "Authorization": "Bearer rowly_live_a_<your-app-key>" }
    }
  }
}
```

## Agent Skill

Rowly publishes a first-party public skill catalog for supported coding agents:

```bash
npx skills add https://rowly.ai --list
npx skills add https://rowly.ai --skill rowly -g -a codex -y
```

- Preferred catalog: `https://rowly.ai/.well-known/agent-skills/index.json`
- Legacy catalog: `https://rowly.ai/.well-known/skills/index.json`
- The public skill contains only first-party Rowly guidance. Tenant-private Skill records stay behind authenticated `/api/v1/skills` endpoints and are not published by default.

## MCP endpoint

- URL: `https://rowly.ai/mcp`
- Transport: Streamable HTTP (MCP spec 2025-06-18)
- Auth: Bearer token in Authorization header
- Tools: current MCP catalog covering workspace, schema, records, workflow, AI memory, Thai accounting, and extensions. Key tools include list_apps, create_record, run_workflow_action, make_payment_from_invoice, and get_thai_pp30_summary.

## Async bulk tasks

- `bulk_update` and `bulk_delete` run synchronously up to 100 matched rows. Larger matches return `{ "async": true, "task_id": "tsk_...", "poll_tool": "get_task_status" }`.
- `export_records` runs synchronously up to 1,000 matched rows. Larger exports write the body to R2 and expose the storage key in `result.storageKey` on the completed task; fetch it through authenticated `/files/get?key=...`.
- Poll `get_task_status` with the returned `task_id` until `status` is `done` or `failed`. Completed tasks include `result`; task ids are scoped to the tenant/app that started them.

## Optional

- [Full docs in plain text](/llms-full.txt): Complete documentation for all pages, suitable for LLM context ingestion.