AI Agent Launch Tools

May 25, 2026 ยท View on GitHub

Small, dependency-free checks for builders launching AI agents, MCP servers, and tool-using workflows.

The first tool is a public launch-surface scanner for sites you own or have permission to test. It checks basic launch hygiene signals before you ship:

  • Security headers
  • Overbroad CORS
  • Cookie flags
  • Source-map exposure hints
  • Public build variable clues

The repo also includes practical MCP/tool-call launch checklists:

It now includes a small config reviewer and tools/list importer. The config reviewer turns a redacted Claude Desktop-style MCP config into a pre-install BLOCK / CAUTION / REVIEW report. The importer turns MCP tool metadata into an allow / ask / deny permission matrix with a snapshot digest for re-reviewing changed tools, without invoking any tools. It also recursively scans tool names, descriptions, and every string inside inputSchema for metadata/schema injection signals, including nested parameter descriptions, enum values, defaults, and titles. It flags schema-quality drift such as missing or empty inputSchema, object schemas without properties, missing required arrays, undocumented parameters, boolean/null/array property-schema entries, union type arrays that need target-client regression coverage, and JSON Schema $ref entries that some MCP clients or LLM tool adapters may not dereference before argument generation. It now also flags missing or incomplete outputSchema metadata for tools that appear to return structured data, so teams can review whether structuredContent can be validated and rendered reliably. It also flags missing or incomplete MCP annotations hints that clients can use for read-only, destructive, idempotent, and open-world approval prompts. It can also print a Codex config.toml review snippet that keeps sandbox settings separate from MCP tool approval.

The repo also includes Supabase launch CLIs for redacted SQL/RPC/view/Security Advisor notes. supabase-rpc-audit checks local text only and flags public-schema definer functions, public views missing security_invoker, broad EXECUTE or SELECT grants, default-EXECUTE revoke mismatches, callable-RPC ACL or REST smoke-test evidence, missing search_path hardening, Function Search Path Mutable review packets, SQL-function inlining tradeoffs, SET search_path FROM CURRENT evidence needs, and privileged functions or views that can bypass caller RLS expectations. supabase-grants-cutover reviews redacted Data API grants and policy packets for the 2026 explicit-grants default, including missing table grants, local supabase db reset / historical migration replay gaps, supabase db pull generated REVOKE replay risk, default privilege state, broad grant quick fixes, function EXECUTE evidence, disabled RLS, permissive policies, anonymous sign-in boundaries, and auth.uid() null behavior. It also extracts redacted PostgREST 42501 grant hints into reviewable GRANT ... statements plus role-matrix smoke tests, so teams can fix missing reachability without turning it into a broad RLS or policy change. supabase-anonymous-rls-audit reviews redacted policies for the anonymous sign-in footgun where temporary users still use the authenticated role, with special attention to invitation, team, membership, billing, owner, profile, and nullable-email policy checks. supabase-tenant-boundary-audit reviews a redacted multi-tenant launch packet for missing tenant boundaries, wrong-tenant negative-test gaps, service-role path mapping, SECURITY DEFINER caller-context tests, Storage upload/upsert policy evidence, broad grants, and billing or owner state transitions. Use --fail-on high in CI to block generated migrations that drop launch-blocking grants, views, anonymous-session boundaries, tenant checks, or RPC safety markers.

The public browser tools also include Supabase launch checks for teams pairing AI agents with Supabase. Use them to review redacted Lovable/v0/Bolt-style generated app packets, Data API grants, explicit grant migration skeletons, multi-tenant RLS boundary packets, anonymous sign-in RLS boundaries, Security Definer RPCs, default EXECUTE exposure packets, security_invoker view drift, exposed views, Security Advisor search_path warnings, auth signup trigger failures, and project-scoped Supabase MCP branching before an agent applies migrations or touches production data. The grants checker now covers the May 30, 2026 new-project Data API default and the October 30, 2026 rollout for existing projects, including default-privilege state and function EXECUTE evidence.

If your immediate problem is moving from Supabase Cloud to a self-hosted Docker stack, use the self-hosted migration trap check before DNS or frontend cutover. It reviews redacted notes for restore files, auth users and JWT/session behavior, Storage transfer, Docker secrets, Data API grants, runtime URLs, and rollback evidence:

https://ai-launch-risk-check-public.vercel.app/supabase-self-hosted-migration-trap-check.html

Use the cutover guide first if you need the source-backed sequence before building a packet. It walks through database restore, Storage object copy, auth/session behavior, runtime configuration, grants/RLS tests, frontend switching, and rollback:

https://ai-launch-risk-check-public.vercel.app/supabase-self-hosted-migration-cutover-guide.html

The fictional sample report shows the expected 24-hour second-pass shape for a redacted Supabase Cloud to self-hosted migration packet:

https://ai-launch-risk-check-public.vercel.app/sample-supabase-self-hosted-migration-report.md

Start with the Lovable Supabase launch preflight checklist if your immediate problem is a Lovable Cloud migration or generated app where auth transfer, RLS, Storage upserts, backend ownership, frontend target switching, or generated migrations are unclear before launch:

https://github.com/kayalopez/ai-agent-launch-tools/blob/main/checklists/lovable-supabase-launch-preflight-checklist.md

For a public migration gate written around the exact Lovable Cloud to owned Supabase cutover:

https://ai-launch-risk-check-public.vercel.app/lovable-cloud-to-supabase-migration-checklist.html

Use the browser preflight to turn redacted migration or launch notes into a packet and migration.md cutover plan. The generated plan marks BLOCK, CAUTION, or REVIEW, then walks through explicit grants, clean supabase db reset replay, role-matrix tests, Storage upsert tests, frontend target switching, fallback, and service-role exposure follow-up:

https://ai-launch-risk-check-public.vercel.app/lovable-supabase-launch-preflight.html

If the concern is that a Lovable-generated app, chat, env name, or frontend bundle may have exposed a Supabase service-role or secret key, start with the local service-role exposure check and keep raw values out of the packet:

https://ai-launch-risk-check-public.vercel.app/lovable-service-role-exposure-check.html

The companion cleanup checklist is here:

https://github.com/kayalopez/ai-agent-launch-tools/blob/main/checklists/lovable-service-role-exposure-cleanup-checklist.md

If that packet needs a 24-hour second pass, the fixed-scope Lovable Cloud migration report page explains the one-packet $25 scope before checkout:

https://ai-launch-risk-check-public.vercel.app/lovable-cloud-migration-risk-report.html

The fictional sample report shows the expected deliverable shape for a redacted Lovable Cloud to owned Supabase migration packet:

https://ai-launch-risk-check-public.vercel.app/sample-lovable-cloud-migration-risk-report.md

Start with the Supabase deadline checklist if your immediate problem is a generated migration, Lovable/v0-style Supabase app, or new project that returns 42501 after a table is created:

https://github.com/kayalopez/ai-agent-launch-tools/blob/main/checklists/supabase-data-api-grants-deadline-checklist.md

For a redacted supabase db pull migration that generated REVOKE blocks and now replays 42501 failures after supabase db reset:

https://ai-launch-risk-check-public.vercel.app/supabase-db-pull-revoke-replay.html

For a PostgREST PGRST204 schema-cache error after adding a column, table, or function, build a redacted packet that separates stale cache from wrong-project, migration replay, grants, and RLS evidence:

https://ai-launch-risk-check-public.vercel.app/supabase-postgrest-schema-cache-pgrst204.html

For a Supabase RPC function ambiguity error such as PGRST202, PGRST203, or PostgreSQL 42725 after changing a function signature, dropping/recreating a function, or using overloads, build a redacted packet that separates request-shape drift, pg_proc signature proof, schema reload evidence, and grants/RLS review:

https://ai-launch-risk-check-public.vercel.app/supabase-rpc-function-ambiguity-packet.html

For a stale Supabase REST response where direct SQL/RLS says rows are deleted or denied but one exact REST URL still returns them, build a redacted packet that separates database truth, URL-key behavior, response headers, client/server cache path, and schema-cache reload evidence:

https://ai-launch-risk-check-public.vercel.app/supabase-rest-stale-response-packet.html

For a redacted generated-app launch packet covering Lovable/v0/Bolt-style migrations, RLS evidence, RPC/function EXECUTE, Storage upload/upsert policies, wrong-tenant tests, and backend ownership:

https://ai-launch-risk-check-public.vercel.app/supabase-generated-app-launch-check.html

For a focused paid handoff, the Supabase Launch Risk Report page explains the one-packet $25 scope and links the free triage tools plus sample report before checkout. It is the right next step when a free checker returns a high or medium finding and you want one 24-hour Markdown report with severity, likely failure mode, and launch smoke tests for a redacted packet:

https://ai-launch-risk-check-public.vercel.app/supabase-launch-risk-report.html

For a redacted default-EXECUTE packet before that handoff:

https://ai-launch-risk-check-public.vercel.app/supabase-rpc-exposure-packet-builder.html

For the 2026 Supabase Data API grants cutover:

https://ai-launch-risk-check-public.vercel.app/supabase-api-grants-readiness.html

For a redacted explicit-grants migration skeleton and role-matrix test packet:

https://ai-launch-risk-check-public.vercel.app/supabase-grant-migration-builder.html

For a redacted multi-tenant RLS boundary packet and wrong-tenant role matrix:

https://ai-launch-risk-check-public.vercel.app/supabase-tenant-boundary-packet.html

Need the full launch workflow? The $25 AI Agent Launch Pack includes the local app, safe-intake builder, checklist, templates, sample report, and optional fixed-scope 24-hour review path:

https://ai-launch-risk-check-public.vercel.app/

Buy Now If

The paid pack is a fit when:

  • You are launching one agent workflow this week.
  • The workflow can read private context or trigger tool calls.
  • You need launch evidence, templates, and a safer intake path today.

Start with the free tools instead when:

  • You cannot describe one workflow without secrets or customer records.
  • You need legal advice, compliance certification, or penetration testing.
  • You only need general reading and the free checklists already cover it.

Free readiness report:

https://ai-launch-risk-check-public.vercel.app/launch-readiness-report.html

If the scope fits, the digital pack checkout starts here:

https://ai-launch-risk-check-public.vercel.app/checkout-after-scope.html#digital-pack

Use

Run the MCP trust verification planner:

npx --package github:kayalopez/ai-agent-launch-tools#v0.1.15 mcp-trust-check --server "candidate MCP server" --workflow "one AI workflow that can read docs and call approved tools"

Review redacted Supabase SQL/RPC/view notes for Security Definer and security-invoker risk:

npx --package github:kayalopez/ai-agent-launch-tools#v0.1.29 supabase-rpc-audit --file supabase_rpc.redacted.sql

Review redacted Supabase Data API grants and RLS policy notes for the 2026 explicit-grants cutover:

npx --package github:kayalopez/ai-agent-launch-tools#v0.1.29 supabase-grants-cutover --file supabase_grants.redacted.sql --fail-on high

Review redacted Supabase anonymous sign-in RLS policies for authenticated-role drift:

npx --package github:kayalopez/ai-agent-launch-tools#v0.1.29 supabase-anonymous-rls-audit --file supabase_anonymous_rls.redacted.sql --fail-on high

Review a redacted Supabase multi-tenant RLS boundary packet:

npx --package github:kayalopez/ai-agent-launch-tools#v0.1.29 supabase-tenant-boundary-audit --file supabase_tenant_boundary.redacted.sql --fail-on high

When either Supabase CLI returns BLOCK or CAUTION, use the generated digest and redacted packet as the intake boundary. Do not send live credentials, connection strings, service-role strings, OAuth material, customer records, payment records, private screenshots, full names, private handles, or full transaction identifiers.

Fail CI on high-severity migration drift:

npx --package github:kayalopez/ai-agent-launch-tools#v0.1.29 supabase-rpc-audit --file supabase_migration.redacted.sql --fail-on high

Try the included Supabase RPC example after cloning:

node scripts/supabase-rpc-audit.mjs --file examples/supabase-security-definer-rpc.sql

Try the included Supabase view example after cloning:

node scripts/supabase-rpc-audit.mjs --file examples/supabase-security-invoker-view.sql

JSON output:

npx --package github:kayalopez/ai-agent-launch-tools#v0.1.29 supabase-rpc-audit --file supabase_rpc.redacted.sql --json

Review a redacted Supabase Security Advisor Function Search Path Mutable tradeoff packet:

node scripts/supabase-rpc-audit.mjs --file examples/supabase-function-search-path-inline.sql

Review a redacted self-hosted/local Supabase hard-coded search-path packet:

node scripts/supabase-rpc-audit.mjs --file examples/supabase-self-hosted-search-path.sql

Review a redacted default EXECUTE exposure packet:

node scripts/supabase-rpc-audit.mjs --file examples/supabase-default-execute-exposure.sql

Review the included redacted Data API grants cutover packet:

node scripts/supabase-grants-cutover.mjs --file examples/supabase-grants-cutover.sql

Review the included supabase db pull generated REVOKE replay packet:

node scripts/supabase-grants-cutover.mjs --file examples/supabase-db-pull-revoke-replay.sql --fail-on high

Extract a redacted PostgREST 42501 grant hint into a review packet:

node scripts/supabase-grants-cutover.mjs --file examples/supabase-42501-grant-hint.txt

JSON output:

npx --package github:kayalopez/ai-agent-launch-tools#v0.1.29 supabase-grants-cutover --file supabase_grants.redacted.sql --json

Try the included anonymous sign-in RLS example after cloning:

node scripts/supabase-anonymous-rls-audit.mjs --file examples/supabase-anonymous-rls-invitation.sql

Try the included multi-tenant RLS boundary example after cloning:

node scripts/supabase-tenant-boundary-audit.mjs --file examples/supabase-tenant-boundary-packet.sql

Review a redacted MCP client config before installing or approving servers:

npx --package github:kayalopez/ai-agent-launch-tools#v0.1.15 mcp-config-risk --file claude_desktop_config.redacted.json

Try the included redacted example after cloning:

node scripts/mcp-config-risk.mjs --file examples/mcp-client-config-redacted.json

JSON output:

npx --package github:kayalopez/ai-agent-launch-tools#v0.1.15 mcp-config-risk --file claude_desktop_config.redacted.json --json

JSON output:

npx --package github:kayalopez/ai-agent-launch-tools#v0.1.15 mcp-trust-check --server "candidate MCP server" --json

Generate an MCP permission matrix from a tools/list JSON result:

npx --package github:kayalopez/ai-agent-launch-tools#v0.1.15 mcp-permission-matrix --file tools-list.json --server "candidate MCP server"

Try the included example after cloning:

node scripts/mcp-permission-matrix.mjs --file examples/mcp-tools-list-example.json

Try the schema-injection example after cloning:

node scripts/mcp-permission-matrix.mjs --file examples/mcp-tools-list-schema-injection-example.json

Try the schema-quality example after cloning:

node scripts/mcp-permission-matrix.mjs --file examples/mcp-tools-list-schema-quality-example.json

Try the $ref schema-risk example after cloning:

node scripts/mcp-permission-matrix.mjs --file examples/mcp-tools-list-schema-ref-example.json

Try the annotations review example after cloning:

node scripts/mcp-permission-matrix.mjs --file examples/mcp-tools-list-annotations-example.json

Try the output-schema review example after cloning:

node scripts/mcp-permission-matrix.mjs --file examples/mcp-tools-list-output-schema-example.json

JSON output:

npx --package github:kayalopez/ai-agent-launch-tools#v0.1.15 mcp-permission-matrix --file tools-list.json --json

Compare against a prior JSON matrix snapshot:

npx --package github:kayalopez/ai-agent-launch-tools#v0.1.15 mcp-permission-matrix --file tools-list.json --baseline previous-matrix.json

Print only a Codex config.toml review snippet:

npx --package github:kayalopez/ai-agent-launch-tools#v0.1.15 mcp-permission-matrix --file tools-list.json --server "candidate MCP server" --codex-config

Run the public launch-surface scanner from GitHub:

npx --package github:kayalopez/ai-agent-launch-tools#v0.1.15 public-surface-scan https://example.com

Or after cloning:

node scripts/public-surface-scan.mjs https://example.com

JSON output:

npx --package github:kayalopez/ai-agent-launch-tools#v0.1.15 public-surface-scan https://example.com --json

Markdown report:

npx --package github:kayalopez/ai-agent-launch-tools#v0.1.15 public-surface-scan https://example.com --markdown

The scanner blocks localhost, private, reserved, and non-standard-port targets. It is for public launch hygiene only, not penetration testing, vulnerability scanning, legal advice, compliance certification, or a security guarantee.

The MCP trust planner blocks obvious secret-like inputs and prints a non-sensitive launch checklist. The MCP config reviewer refuses likely unredacted secret env values, prints env key names but not values, and never starts servers, calls tools, fetches URLs, or validates safety. The MCP permission matrix importer reads metadata only, never invokes MCP tools, recursively scans nested schema strings for prompt-injection and tool-selection-bias signals, flags schema-quality drift that can make clients lose argument metadata, and now highlights $ref-based schemas, non-object property schemas, union type arrays, missing/incomplete outputSchema metadata for structured results, and missing/incomplete MCP annotations hints that may need target-client regression coverage before launch. It can compare a new metadata snapshot against a prior JSON matrix so added, removed, or changed tools re-enter review. The Codex snippet maps allow to approval_mode = "approve", ask maps to approval_mode = "prompt", and deny maps to disabled_tools; it does not disable sandboxing. The Supabase CLIs read local redacted text only; they do not connect to Supabase, fetch URLs, call RPCs, inspect private data, or prove safety. These tools are for review planning only, not approval automation, penetration testing, compliance certification, or a guarantee that a server or database is safe.

Free Browser Tools

These no-login tools are live:

The full paid pack and fixed-scope review are described on the product page:

https://ai-launch-risk-check-public.vercel.app/

Safe Use

Only scan public URLs you own or have permission to assess. Do not paste secrets, tokens, private customer data, cookies, payment pages, internal hosts, or non-public endpoints into this tool.