SaaS Pro Max
Sign inStart free
Documentation menu

Set up a feature with your agent

First-time setup pages include a feature preview, up to three setup steps and Set up with your coding agent. Creation pages offer the same assistance above the form. The preview is illustrative, not telemetry from your app.

Copy a scoped prompt#

  1. Open the feature in the correct application and environment.
  2. Choose View prompt to inspect the complete instructions and optionally name your framework, or Copy setup prompt to let the agent detect it.
  3. Paste into your coding agent with the application's repository open.

The prompt includes organization and application identifiers, the selected environment, API origin, feature-specific instructions and documentation. It asks the agent to inspect existing code, preserve consent and permission checks, prepare changes and verify actual receipts. Copying is local; SaaS Pro does not send the prompt to an agent automatically.

No stored credentials are added. A public key created on the Install page is available once, in that page's memory. View prompt lets you explicitly include it before copying. Leaving or reloading loses that value. Public keys can collect events and evaluate flags; they cannot manage the application. Secret keys, provider credentials and management tokens are never added to the generated prompt. Existing key prefixes cannot be reconstructed into a full key.

Connect Claude or another MCP client#

Open Connect your coding agent or Account → Coding agents and MCP. The published SaaS Pro Max Claude connector uses Google, your passkey and an organization-scoped consent screen. No token needs to be copied. Start with read-only for reports and approve write access only for the changes you need.

Connect agent on setup pages offers Claude Code over hosted MCP and the server address for other remote clients that support OAuth with automatic client registration. Claude Code completes sign-in through /mcp.

Connect local MCP#

Choose Connect agent for Codex or Cursor local configuration. The existing @saaspro/mcp package uses stdio and calls the Management API with SPM_TOKEN and SPM_HOST.

Set the token in your agent's protected environment or local MCP configuration. Keep credential-bearing config files out of Git. Use read scope for inspection, write for reviewed module changes, and admin only when public or secret key operations are required. Scopes do not grant permissions your membership lacks.

MCP has dedicated tools for the core workflows. Features without a dedicated tool can use a documented Management API endpoint. The prompt tells the agent to inspect the deployment's /api/v1/openapi.json before making requests and to leave console-only actions as explicit operator steps.

Cloud agents and ChatGPT#

An agent with an HTTPS API client can use the Management API from its protected server environment. Store SPM_TOKEN in that agent's secret store. A cloud agent cannot reach a localhost deployment; use the origin of the deployed instance you intend to access.

The hosted endpoint is https://saaspro.dev/api/mcp. Remote MCP clients must support Streamable HTTP and OAuth with dynamic client registration and PKCE. Client availability and configuration vary; a Claude directory listing does not imply a listing in another provider's directory. Local stdio configuration is also available when a client does not support the hosted authentication flow.

Data sources and exports#

Some setup tasks are completed in the console rather than a public API:

  • PostgreSQL sources: create a read-only reporting source in Data integrations → Sources. Review the reporting table, tenant column and value, identity binding and allowed columns. The password, CA and join key belong in the vault. Read a bounded preview before approving scheduled snapshots.
  • Storage exports: create an S3-compatible export in Data integrations → Exports. Review bucket, dedicated prefix, credential scope, selected events, bucket lifecycle and schedule. Approve data transfer before enabling delivery, then inspect its receipt and object.

The agent can prepare integration code and a configuration checklist without receiving these secrets in its prompt. It must not invent API routes for actions that the deployment only exposes in the console.

Finish with evidence#

A prompt being copied, a successful SDK install or a passing build is not a collection receipt. Review the changed files, complete any required console steps and verify the first real event, recording, trace, import or delivery in the selected environment. Production writes, external messages, paid runs and deployment remain deliberate operator decisions.

Run the setup wizard#

With the project's repository open, an agent can run either command without terminal prompts:

Shell
npx @saaspro/wizard
curl -fsSL https://saaspro.dev/wizard.mjs | node --input-type=module

A person compares the terminal code at /device, signs in with a passkey and approves the organization, application and environment. Approval creates a public ingest key and requires keys.write plus a fresh passkey. The CLI keeps its RSA private key in memory and decrypts the one-time encrypted reveal.

Use --dry-run to preview planned files and commands (approval still runs unless --key supplies an existing public key). The wizard writes the framework's public key variable in .env.local or .env, adds that filename to .gitignore, creates absent initialization files and wires a single unambiguous React root, Vue mount entry or Astro layout. It supports Next.js App/Pages Router, Remix/React Router, Vite/React, SvelteKit, Nuxt, Vue and Astro, with manual setup for plain Svelte, mobile adapters and existing or ambiguous source. It also merges hosted MCP into .mcp.json and .cursor/mcp.json. Conflicting values cause a refusal. Without a detected consent manager it uses the SDK's granted default; a recognized manager generates pending consent with exact grant/revoke calls to connect to the manager. Use --consent pending or --consent granted to override detection. Nuxt requires the printed runtime-config snippet. It prints remaining initialization and adapter steps, then sends spm.wizard.verified and checks acceptance. Python initialization needs a separately managed secret key; the wizard never creates or writes one. MCP requires separate browser OAuth sign-in. See the wizard guide for flags, exact filenames and verification boundaries.