Start Here

Choose the right first path for Spala: dashboard builder, frontend integration, AI agent, or raw MCP.

Start Here

Spala is a hosted backend builder for AI-built apps. Use the hosted dashboard to create and publish project APIs. Connect AI coding agents through Spala MCP when an agent needs to discover projects, read context, or hand off into a project-specific MCP.

Pick the path that matches your job.

AI agents install Spala first

Start immediately even when the app workspace is empty; do not ask for repository files before connecting Spala. From the app workspace, run pnpm dlx @spala-ai/mcp-install init --client <client> --yes --json and execute the returned setup steps. Run pnpm dlx @spala-ai/mcp-install status --json. If Public MCP cannot hot-load, restart or resume the same session once; do not create a project manually. Complete browser sign-in or signup if requested. For a new account, answer one concise terminal question for the exact missing profile/company fields; the agent calls account_setup. Project naming is the next, separate question before the agent creates or reuses the project. Do not send the user to the dashboard for workspace or project creation and do not start a separate local-only backend.

Boundary glossary

| Term | Meaning |
| --- | --- |
| Spala platform | The hosted dashboard, builder runtime, platform auth, updates, and managed infrastructure operated by Spala |
| Project | The backend app you build in Spala: tables, functions, endpoints, tasks, triggers, channels, addons, settings, docs, and frontend assets |
| Publish backend API | Promote the project draft to the live generated API that frontends and integrations call |
| Deploy generated runtime | Advanced CI/CD path for deploying a generated project runtime outside the default hosted publish flow |
| Frontend App | Static UI assets hosted beside the generated project backend |
| Project Agent | Backend automation that belongs to one Spala project and runs from chat, schedules, webhooks, data changes, or realtime events |
| AI/MCP Agent | External coding agent or MCP client, such as Codex, Claude, Cursor, or VS Code, that helps build or inspect a project |
| Public MCP | https://mcp.spala.ai/mcp, used for discovery, auth metadata, docs search, project lookup, and handoff |
| Project MCP | The project-scoped MCP URL returned after auth and project selection; backend changes happen there |

Before you build

Start from the hosted dashboard at https://dashboard.spala.ai. New users should create an account from the dashboard sign-up screen. If you are joining an existing company workspace, accept the workspace invite or ask the workspace owner to add you. You also need permission to create or edit a project. A first Copilot-built API usually takes a few minutes; at the end you should have tested endpoints, a published API base URL, and generated API docs for frontend work.

Delivery actions

| Action | What changes | Who manages it | Where it lives |
| --- | --- | --- | --- |
| Publish backend API | Draft project resources become the live API | You approve publish; Spala serves the hosted project API | Project API base URL |
| Publish frontend app | Static UI assets become the live project UI | You upload or agent-build the frontend; Spala hosts the assets | Project frontend URL |
| Deploy generated runtime | Generated project runtime is sent to an external target | You configure the target and credentials | Your server or deploy provider |
| Use project MCP | An authenticated agent edits one selected project | You grant access; agent uses returned MCP URL | Project-scoped MCP endpoint |

I am building in the dashboard

Use this path when you want to create a backend from a prompt and publish it from Spala.

1. Open https://dashboard.spala.ai.
2. Create an account, sign in, or accept your workspace invite.
3. Create or open a project.
4. Start in Lite Mode from the project header or mode switch.
5. Ask AI Copilot for the backend feature you need.
6. Review generated tables, functions, endpoints, tasks, triggers, channels, and settings.
7. Test in API Playground.
8. Use Publish backend API to make the generated project API live.

Quick Start
Build and publish a Cases API from the dashboard.
/getting-started/quickstart

AI Copilot
Describe the backend feature and review generated resources.
/features/ai-assistant

Lite Mode
Use the Copilot workspace and project graph as the first screen.
/features/lite-mode

I am connecting a frontend

Use this path when you already have, or are building, a React, Next.js, Vite, Webflow, mobile, or external frontend.

1. Find the project API base URL in Lite Mode or Project Overview.
2. Open Settings → API Docs for the generated Markdown, JSON, OpenAPI, or SDK contract.
3. Find routes in Endpoints or the generated API Docs.
4. Use /api/... route paths exactly as published.
5. For protected endpoints, send Authorization: Bearer USER_JWT.
6. Store environment values in your frontend as VITE_SPALA_API_BASE_URL or NEXT_PUBLIC_SPALA_API_BASE_URL.
7. Test calls in API Playground before wiring the frontend.

Frontend Integration Cookbook
Copyable examples for base URLs, auth, uploads, realtime, and env vars.
/guides/frontend-integration

External Frontend Handoff
Share the exact packet a frontend developer or AI agent needs without dashboard access.
/guides/frontend-handoff

API Endpoints
Create and inspect REST API routes.
/endpoints

API Docs
Export OpenAPI, Markdown docs, SDK files, and snippets.
/features/api-docs

I am an AI coding agent

Use this path when Codex, Claude Code, Cursor, VS Code, or another agent needs to work with Spala.

1. Start even if the app workspace has no files. Run pnpm dlx @spala-ai/mcp-install init --client <client> --yes --json from that workspace and execute every setup step it returns.
2. Run pnpm dlx @spala-ai/mcp-install status --json.
3. Complete browser sign-in or signup only when requested.
4. Call spala_get_onboarding, spala_get_tool_map, then account_status on Public MCP.
5. If setup is required, ask once for exactly missingFields and call account_setup with real values. This creates the missing account workspace; do not send the user to dashboard onboarding.
6. As the next distinct step, ask for or derive the project name. Reuse .spala/project.json, list projects, or call project_create only when needed.
7. Call project_connect exactly once with the chosen project ID.
8. Execute the returned workspace-only bind plan. Do not install a project MCP globally.
9. Continue on the connected project MCP and inspect project context before making changes.
10. Preview, validate, apply, publish backend API when requested, and review through the project MCP.

Do not guess project MCP URLs

Project MCP URLs are resolved after authentication from Spala project data. Do not hardcode https://api.spala.ai/{{project}}/mcp or any other pattern.

Spala Public MCP
Public MCP tools, OAuth metadata, raw transcript, and project handoff.
/agents/mcp

Codex Setup
Add Spala Public MCP to Codex.
/agents/codex

Cursor Setup
Configure Cursor with the public MCP entrypoint.
/agents/cursor

More MCP Clients
Gemini CLI, VS Code/Copilot, Windsurf, and Cline setup paths.
/agents

Public Agent Skills
Skill files that help agents evaluate Spala and start safely.
/agents/skills

I am implementing raw MCP

Use this path only if you are building an MCP client or debugging MCP transport directly.

1. Use POST https://mcp.spala.ai/mcp.
2. Send Content-Type: application/json.
3. Send Accept: application/json, text/event-stream.
4. Call MCP protocol methods such as initialize, tools/list, and tools/call.
5. Pass Spala tool names inside tools/call; do not call tool names as top-level JSON-RPC methods.
6. Discover OAuth endpoints from MCP OAuth metadata.

Spala Public MCP
/agents/mcp
Full raw MCP notes and transcript
Project Delivery
/deployment
Publish projects without self-hosting Spala itself
Settings
/settings
Configure API docs, auth, env vars, MCP, and publishing