Supero MCP Server

SECURITY + CORRECTNESS AUDIT of a bundle BEFORE go-live (static; your code is never executed). Finds what build_validate/build_doctor do not: MONEY-01 client-chosen checkout amounts, AUTHZ-01 end-users able to write server-owned fields (paid/issued/status/amount/payment refs) or create tickets/payments, FLOW-01 fulfilment done in the browser or on payment_status alone, INV-01 capacity/stock with no server-side enforcement, PUB-01 draft/hidden rows in PUBLIC_SCHEMAS, LAUNCH-01 terms/privacy/password-reset/meta gaps. Each finding has severity (blocker|critical|high|warning), file:line or policy, evidence, the exact fix pattern and a fingerprint. Fix every blocker and critical, then re-run until clean; build_go_live reports (or, when enforced, refuses on) outstanding blocker/critical findings. Pass files/files_ref, or only project_uuid to audit the latest published version.

Bind source streams to the app's schemas. mode='live_read' (default: read-only BYODB / warehouse), 'live_readwrite' (BYODB read+write, DOMAIN-ADMIN only), or 'sync' (copy into SuperoDB). Modes by source: postgresql, supabase, mysql take all three; mssql takes sync and live_read; oracle and mongodb take sync only; a REST api takes sync and live_read; warehouses take live_read only. For live modes, discovery must have run and each binding's `source` must match a discovered stream; the bind VERIFIES the live mapping materialized. A NEW live table must state tenant_binding: ask the data owner which rows belong to which tenant, never assume. Bound live schemas are read/written via ORDINARY app CRUD, nothing goes in the bundle.

Configure a service's keys for the project so an integration works at deploy (e.g. wire SendGrid for email, a Stripe TEST key for checkout). TEST/SANDBOX keys ONLY. Keys with a clear live marker (Stripe sk_live_… / Razorpay rzp_live_…) are auto-refused, but most providers give NO test-vs-live signal — so for EVERY provider send test/sandbox keys only and use the admin panel for production secrets (a deep link is returned). Never put a real secret through this tool/chat. Reacts to the live permission result; secret values are never echoed.

Create a data connector to an EXTERNAL source the app owner controls (their own database, a REST API, or a warehouse). Returns a connector_id. Does NOT bind schemas yet. Flow: build_connect_data_source → build_discover_source (lists the tables/streams; needs this connector_id) → build_bind_data_source (bind streams to app schemas) → build_run_data_source (mode='sync' bindings only; live reads need no run). For a database you do NOT need to know the tables upfront: omit `table_names` and let discovery + bind pick them (the connector is a manual-trigger shell until bound, and a run before a sync bind is refused). Pass `table_names` only when you already know exactly which tables to sync. Credentials are sent to the platform and NEVER echoed back; prefer read-only DB creds / the Key Store for production. Requires a domain- or project-admin API key.

Create a NEW project (domain-admin keys only; plan-gated). Mints the project's schema_namespace + an API key (returned ONCE). Use for a fresh app.

Deploy a published version. target='cloud_ephemeral' (managed Cloud Run, ~30m throwaway preview; requires platform enablement; the default when cloud deploy is enabled) or 'local' (hand the user the bundle to run with the project's own key). For a PERMANENT public URL, use build_go_live instead.

Poll a cloud deploy started by build_deploy or build_go_live. Pass the poll_url it returned. Reports elapsed_s since launch. 'launching' for 5-6.5 minutes is normal (Cloud Run IAM propagation): keep polling every ~30s and probe the url once it appears. Only a persistent unreachable past ~10 minutes is a real failure (see the anti-loop rules).

Discover a connector's schema: trigger discovery, wait, and return the source streams (names, columns, primary keys), AI-inferred Supero schemas, field mappings, and a suggested namespace. Run AFTER build_connect_data_source and BEFORE build_bind_data_source (a live bind's `source` MUST equal a discovered stream name).

PREFLIGHT a bundle BEFORE publish/deploy — catches the silent deploy-killers build_validate does NOT: missing #supero-preloader removal (app stuck on a spinner forever), heavy startup seed (Cloud Run port-bind timeout → 'container failed to start'), reserved field names like status/state (silently dropped), namespace collisions (ambiguous reads), and services needing elevated import permission. Run it after build_validate and before build_publish.

Fetch the CURRENT (or a given) PUBLISHED bundle: the file list + a signed download_url, or ONE file's content inline via file=. For ANY change request on an existing app, START from this bundle and modify it — re-authoring from scratch silently drops the hand-authored ui/app.js and every prior fix.

Fetch a COMPLETE, production-quality reference app (schemas.py + config.py + setup.py + ui/app.js) to copy patterns from — the single biggest lever for app quality. archetype='index' (default) lists the available archetypes with per-file byte sizes; pick the one closest to your app ('commerce-marketplace' | 'service-booking' | 'ops-dashboard' | 'multitenant-portal' | 'saas-billing') and fetch it BEFORE authoring your UI. These are real 'stunning' apps; mirror how they compose the SDK components, art-direct the landing page, and structure schemas. Optional file= to fetch just one file ('schemas.py' | 'config.py' | 'setup.py' | 'ui/app.js'). Responses are PAGED (~32KB/section): when the header says more:true, fetch the next section with offset=<next_offset> — never re-fetch from 0.

Get one project's details: schema_namespace (use this EXACT value as the `namespace` literal on every schema dict), last published version, AND the captured `project_intent` — the brief (project_description/summary), the discovered data model (entities/central_entity/relationships/status_workflows), and the landing intent (public_landing_view). BUILD TO THIS — it's the authoritative app spec the user already gave; don't re-ask or ignore it.

Fetch the AUTHORITATIVE contract for a transactional/stateful platform service (cart, order, payment, booking, appointment, membership, approval, document_signature, recurring_plan, inventory, task, ticket, loyalty_points, rental, comment, attachment, feedback, notification, product, service, customer, workflows). Returns the service's state machine (initial_state + transitions), its operations (op ids + input fields + resulting state), the base schemas + mandatory fields you must supply, AND the platform's DEFAULT UI SCHEMAS for that service — the bulletproof reference for building a correct, sophisticated transactional UI. ALWAYS call this for any service your app `extends` BEFORE authoring its UI — do not guess op names, states, or mandatory fields from prose. service_id='index' (default) lists all services.

Fetch a Supero build reference. doc='skills' (default, the spec you MUST follow) | 'components' (the real pre-built SDK component/global catalog — read so you don't reinvent UI) | 'rubric' (the rich-app quality checklist — what 'stunning' means) | 'landing' (compact landing-page derivation + quality bar — MANDATORY read for public-facing apps) | 'integrations' (the EXACT services.* wrapper→service_id→args map for email/sms/stripe/ai/etc.) | 'web' / 'transactional' / 'workflows' / 'mobile' / 'services' (deep companion docs) | 'e2e_testing'. Returns the doc + a content version + the SDK floor to pin. Read 'skills' FIRST, then 'components' + a matching build_get_examples before you author UI. Docs are PAGED (~32KB/section): when the header says more:true, fetch the next section with offset=<next_offset> — never re-fetch from 0.

PERMANENT deploy: promote a published version to a PERMANENT public URL at <service>.supero.live (managed Cloud Run) — unlike build_deploy(target='cloud_ephemeral'), which is a ~30-min throwaway. If version_uuid/file_id are omitted, the latest published version for the project is used. Returns public_url + poll_url; poll with build_deploy_status until live, then build_smoke_test the public_url. build_teardown removes it. Requires a domain- or project-admin API key + platform cloud-deploy enablement.

IMPORT platform services into a project (e.g. 'workflows', 'email', 'stripe_checkout', 'cart', 'order', 'payment') so the app can call them and its event-bound workflows can run. Uses the platform's service import with YOUR connection's credential. Reports each service's exact outcome: imported, already imported, or FAILED with the platform's reason (schema_drift, schemas_failed, error_code, HTTP status) — a failed import is never reported as success, and you must not ship an app whose required service failed to import. After importing, set keys with build_configure_services. MCP-IMPORT-SERVICES-V1.

Classify this project's schemas: which are connector-backed vs app-authored, and each one's access mode (sync | live-ro | live-rw | warehouse). Use it so you DON'T render create/edit UI for read-only live sources (live-ro/warehouse) or regenerate/overwrite connector-discovered schemas.

List the platform's available services/integrations from the LIVE catalog (email, sms, slack, ai, stripe_checkout, google_oauth, push_notification, …) — so the intake's 'which connections/integrations?' question is accurate and you never guess a service id. Returns each service's exact catalog `service_id` (use it verbatim in config.py `services` — e.g. 'stripe_checkout', NOT 'stripe'), category, whether it needs a key, and YOUR connection's service permissions (can_import / can_configure). Descriptive — what EXISTS, never what to use. Pass service_id for one service's config fields.

List the external DATA SOURCE types an app can connect to — its own Postgres/MySQL/MSSQL/Oracle/MongoDB, any REST API, or a Snowflake/BigQuery/Redshift/Databricks/ClickHouse/Fabric warehouse — plus the curated public-API catalog. Read this to offer a 'connect your own data' option. Flow: build_connect_data_source → build_discover_source → build_bind_data_source. See build_get_skills(doc='connectors').

List projects you can build into (uuid, name, schema_namespace, live_url).

Fetch recent Cloud Run logs for a project's deployed app — THE tool for diagnosing a failed/stalled cloud deploy or a crashing app (startup-probe timeouts, tracebacks, 'container failed to start on PORT'). Read-only; rate-limited per domain.

PLAN FIRST — turn a one-line app idea into an explicit BUILD CHECKLIST before you author anything, so a thin prompt doesn't silently skip what expert builders add (this is exactly why first-draft apps miss detail pages, tenant pickers, seed data). Deterministic, no LLM: it detects the app's VERTICAL and returns the authentic page structure + terminology for that domain, which entities need a full DETAIL PAGE, the multi-tenant login pattern (picker + tenant=''), seed guidance, and which build_get_examples to copy. Call it right after build_get_skills and BEFORE authoring.

Package + upload your authored bundle and record a version under the project. Returns version_uuid + file_id + download_url. Runs build_validate first; force=true overrides soft errors only. Provide files as {relative_path: content}.

RECOMMEND which concrete provider integrations this app needs, and WHY — the deterministic Step-2 intelligence the web wizard uses, now over MCP. Pass the app `description` + the platform `service_ids` it will use (e.g. ['cart','order']); returns GROUPED, TIERED suggestions (required/recommended/optional) with the default option flagged — e.g. cart/checkout → a payment gateway (REQUIRED; stripe_checkout default, paypal/razorpay offered), customer-facing apps → transactional email, appointments+reminders → sms. Options are drawn ONLY from the LIVE installed manifests, so it can't suggest a provider you don't have. PRESCRIPTIVE complement to build_list_capabilities (which is descriptive): call this so you don't OMIT a needed integration; use build_list_capabilities for a service's exact id + config fields.

DESTRUCTIVE (DOMAIN-admin keys only): completely replace a DEV-mode project's app — wipes its data (retains credentials, API keys, namespace, tenant). Refused if mode='live'. Requires confirm_project_name to match. If you pass files, they are VALIDATED before any wipe (a bad bundle is a no-op) and published after; otherwise wipe-only, then call build_publish.

Trigger a SYNC RUN on a data connector and return the run id. MCP-created connectors are trigger:manual, so a mode='sync' binding copies NO rows until a run executes — call this after build_bind_data_source (and again whenever the source data changes). Live-read/warehouse bindings don't need runs (they read the source directly).

Set a project's build mode. 'dev' (default) allows COMPLETE REPLACE (wipe data, keep credentials); 'live' protects it. Switch to dev before replacing, to live when it's in production.

VERIFY a DEPLOYED app actually works (not just 'running'). HTTP-checks the live URL: root loads with a title, app.js is your bundle (not a stub) and dismisses the boot splash, config.js namespace matches the project; optionally logs in and reads an entity to confirm data + no namespace ambiguity. Pass url= (from build_deploy_status) or poll_url=. THE post-deploy gate — run it after every deploy.

PUB-1a: get the out-of-band UPLOAD endpoint for a LARGE app bundle that won't fit inline (the model's max OUTPUT tokens cap `files`/`files_b64gz`, so big apps otherwise have to be truncated/minified). Upload a gzip(json {path:content}) blob to the returned URL with your OWN key, then pass the returned file_id as `files_ref` to build_validate / build_publish / build_doctor. Bundle size then no longer depends on any token cap.

Tear down a project's live deployment — BOTH the permanent (build_go_live) app and the ephemeral preview — DELETEing the managed Cloud Run services and freeing their URLs. Idempotent: a project with nothing deployed returns stopped=true. ONLY manages Supero-hosted apps — a project deployed to your own AWS/GCP is refused, not silently reported stopped. Check `stopped`: false means the teardown was INCOMPLETE and the app may still be serving (and billing) — re-run it, do not report success. Requires a domain- or project-admin API key.

Update a project's metadata (display_name, description, show_public, live_url).

Validate a locally-authored bundle against the live platform BEFORE publishing. AST-only (your code is never executed). Checks manifest, syntax, import-safety, config exports, schema validity, and namespace==project schema_namespace.

ACCEPTANCE TEST of your app's running PREVIEW (after build_deploy, before build_go_live). Where build_audit reads the code, build_verify USES the deployed preview like a buyer would, with throwaway test accounts (verify-<rand>@example.com, deleted afterwards): health (serves, app.js = the published version, boots, API reachable); accounts (sign-up, login right after sign-up, login, forgot-password reachable); the role permission matrix (per end-user role x schema: the declared policy holds, server-owned fields such as status/totals/payment state cannot be written by the user, fulfilment records cannot be minted, another user's records cannot be read); commerce (the server prices checkout even when the client price/amount is tampered; a success-URL visit without payment changes nothing; Stripe TEST mode only, otherwise refused); event-bound workflows reach COMPLETED with the verified recipient (no real email is ever sent); capacity is enforced; draft/hidden rows stay out of public reads; launch basics (terms/privacy, meta, viewport, gzip, headers). Returns a score and pass/fail/skip per check with evidence and the exact fix. Only the caller's own project PREVIEW, never the live app or other projects. ~1-3 min, rate-limited; dry_run=true lists the planned checks. Flow: build_audit -> build_publish -> build_deploy -> build_verify -> fix -> build_go_live (which reports the latest build_verify result).

Resolve your API key's scope: role (domain_admin/project_admin), domain, whether you can create projects, and your plan. Call this to decide the flow.

Cancel a running connector execution.

Disable a connector.

Trigger schema/metadata discovery for a connector.

Get discovery results.

Get discovery execution status.

Enable a connector for scheduling.

List available connector plugins.

Trigger a manual connector sync execution.

Get connector status and recent executions.

Test an existing connector's connectivity.

Test a connector configuration.

Delete a schema from the domain. Use with caution.

Get full details of a specific schema by UUID. Returns the complete schema definition including all attributes and metadata.

List all existing schemas in the current domain.

List schemas linked to the current project.

Save/upload schemas into a project. Schemas are validated before saving and linked to the project. Returns the saved schemas with their UUIDs; skipped schemas carry the platform's reason. IMPORTANT: Pass FLAT schema objects directly — do NOT wrap in schema_content. The tool handles schema_type detection and wrapping automatically. Required fields: name, namespace, parent_type, attributes. Set namespace to your app's namespace (e.g. 'shop'); the schema is stored as namespace:name. Mark required attributes with "mandatory": true. Supported attribute types: 'string' for text, 'float' for decimal numbers (NOT 'number'), 'integer' for whole numbers, 'boolean' for true/false. project_uuid defaults to your key's project; a project-scoped key can only save into its own project. A schema skipped for compatibility warnings is applied with force_upload=true. Example: { "schemas": [{ "name": "product", "namespace": "shop", "description": "Product catalog entry", "parent_type": "tenant", "attributes": [ {"name": "title", "type": "string", "mandatory": true}, {"name": "price", "type": "float"}, {"name": "stock", "type": "integer"} ] }] }

Update an existing schema definition.

Validate schemas before saving. Checks for conflicts with existing schemas, circular dependencies, reserved names, and parent_type correctness. Always validate before saving!

Create a new object in the current domain. Use schema_list to see available types and their fields.

Delete an object from the current domain.

Get a specific object by UUID from the current domain.

Search/list objects of any type in the current domain. Use schema_list first to discover available object types. System types: project, tenant, user_account, api_key, connector, audit_log, schema_registry, client_sdk, connector_execution, connector_plugin. Domain types vary per domain (e.g. customer, invoice, appointment, vehicle).

Update an existing object in the current domain.

Get download URL for an SDK.

Generate client SDKs for the domain's schemas.

List available SDKs for the current domain.

Check SDK generation request status.

Check if the current user has a specific permission.

Get current user's role, permissions, and scope.

Get the scope and permissions of the current API key.

Link schemas to a project. Accepts schema UUIDs or names.