Key concepts¶

The words Transica uses, in the order you meet them.

Your account¶

Organization: Everything you own (sites, connectors, tests, plan and credits) belongs to an organization. Every person has a personal organization of one; a team shares an organization with roles: admin, member or viewer. See Roles and permissions.
Site: One website, identified by its address (for example https://example.com). Tests, connectors, hosting and listings all live under a site. Your plan sets how many sites you can have.
Verification: Proof that you control the site's domain: a meta tag on the home page, a file at /.well-known/transica-verify.txt, or a DNS TXT record. Agent tests only run on verified sites, and verification is re-checked before each test. See Add and verify your site.
Test account: A dedicated login on your site (never a real customer's) that agents use to test pages behind sign-in. A site can have up to 10, one of them the default. Passwords are stored encrypted and never shown again.

Task: Something a customer would ask an assistant to do, in their words: "Find the price of the M8 hex bolt pack". A test runs up to 3 tasks.
Success check: How a run is graded: Answer mentions… (the agent's answer must contain a value), Reaches page…, Page shows text…, or none (the agent reports for itself). Comparisons need an answer check.
AI browser test: An AI agent does each task on your website in a real browser, 1, 3 or 5 times per task. Signed out it's read-only; with a test account it can change data you allow.
AI connector test: An AI agent does each task using only your connector's tools, with no browser.
Comparison: The same tasks, the same AI model and the same limits run through your website and your connector, several trials each. The report says how much faster (or not) the connector is, with evidence.
Explore: A swarm of read-only AI agents that maps your site area by area and records what it can do and which API calls it makes. Explores are named runs and are the easiest way to find what to build.
Task set: A saved, named list of up to 10 tasks with their checks, reused across tests.

Connector: A versioned set of tools that AI apps can call. Built from an API spec, from tests and explores, or by connecting an MCP server you already run.
Endpoint: One operation of your API (GET /api/plans). You tick the endpoints that should become tools; each ticked endpoint becomes one tool. Fewer is better: aim for under 25, up to 50.
Tool: What an AI app sees: a name (get_subscription_plans), a title, a description, inputs and hints (Read-only, Destructive, Open world). Tools are marked Reads, Writes (changes data) or high risk (never runs).
Design: The AI pass that turns ticked endpoints into tool names and descriptions. Each design saves a new draft version for you to review. Designs are counted per month by plan.
Version: A snapshot of a connector's tools. A version is a draft until you publish it; published versions are kept forever and can be viewed any time. Editing a published version creates the next version.

Hosting: Transica serves a published connector over MCP at its own permanent address.
Address: Each hosted connector has one address made from its short name: https://yourapp-mcp.transica.dev/mcp. The short name locks while a directory submission is open.
Own domain: On Business and up, the connector can answer at https://mcp.yourdomain.com/mcp with a branded sign-in page. You prove the domain with a DNS record; our address keeps working alongside it.
AI agent discovery: An optional switch on Hosting that publishes small files telling visiting AI agents where your MCP server is. Off until you turn it on.
Live app: One hosted connector at one address. Plans are priced per live app; drafts, tests and new versions don't count.
Sign-in provider: The service your site already uses for logins (Supabase, Clerk, Auth0, WorkOS, Descope, Stytch, Better Auth, Logto, Keycloak or any OpenID Connect provider). Hosting sends people through it so tools act as them.
Per-tool sign-in: Each tool either needs a login or not. Public tools answer signed out; the first call that needs a login asks the person to sign in.
Write approval: Tools that change data only run once an admin approves them for hosting. High-risk actions (payments, account closure, credentials, messaging others) are never recommended or run.

Listing: Your app's entry in the ChatGPT app directory or the Claude connectors directory: name, description, logo, links, example prompts, reviewer account.
Directory checks: Transica's rule-by-rule check of your connector and listing against each directory's published requirements. Results are ready, needs you or fix first, never approval.
Test prompts: Prompts that should (positive) or shouldn't (negative) make an assistant call your tools. ChatGPT asks for exactly 5 positive and 3 negative; Claude for at least 3.
Rules-change alerts: We re-read each directory's published rules every day. When a rule changes, the checks are updated after our review, and on Business and up admins of affected apps get an email.

Agent traffic: Which AI agents and crawlers visit your website, from access logs you upload or records a small Cloudflare Worker sends, and, once you host, the tool calls your connector gets.
Test credits: What AI-heavy tests cost. One credit is one website run on the standard model; advanced models cost 3, an explore agent run 0.3, and connector-only runs are free. See Plans and credits.

Get startedWhat Transica does Get startedQuickstart Product tourConnectors

Updated 26 Oct 2026