Glossary¶
Terms used across Transica and the AI app world, A to Z. For the core product ideas in reading order, see Key concepts.
- Agent standards
- A separate, optional score in the AI readiness scan for agent-specific files and protocols, such as Markdown for agents, Content Signals, an API catalog and an MCP server card. It never changes the readiness score.
- Agent traffic
- The AI agents and crawlers that visit your website, read from access logs you upload or from a Cloudflare Worker you add, and your hosted connector's tool calls.
- AI agent
- An AI model that takes steps on its own, such as clicking through a website or calling tools, to finish a task.
- AI app
- An assistant people use, such as ChatGPT, Claude, Codex or Cursor, that can connect to MCP servers.
- AI browser test
- A test in which an AI agent does tasks on your website in a real browser.
- AI connector test
- A test in which an AI agent does tasks using only a connector's tools.
- AI catalog, MCP server card
- Small files that tell visiting AI agents your site has an MCP server and where. Served by hosting when you turn on AI agent discovery.
- Comparison
- The same tasks through your website and your connector, with the same model and limits, several trials each.
- Connector
- Transica's name for an MCP server you build, connect or host here: a versioned set of tools.
- Consent screen
- The page where a person approves an AI app's access to their account on your connector, before your own login.
- Content Signals
- Lines in robots.txt that say how AI may use your content: for search, as input to answers, or for training.
- Design
- The AI pass that turns ticked endpoints into tools with names, titles, descriptions and hints.
- Destructive hint, open-world hint, read-only hint
- MCP tool annotations. Read-only: the tool changes nothing. Destructive: it may delete or overwrite. Open world: it reaches outside your own system. Directories ask for them.
- Directory
- An AI app's catalogue of listed apps: the ChatGPT app directory (apps are called plugins) and the Claude connectors directory.
- Endpoint
- One operation of your API, such as
GET /api/plans. - Explore
- A swarm of read-only AI agents that maps your site and records the API calls its pages make.
- Exposure
- Controls hidden from people but reachable by agents. Shown in scans and website tests; never scored.
- Issuer URL
- Your sign-in provider's base address for OpenID Connect, used to find its configuration.
- Live app
- One hosted connector at one address. What paid plans are priced by.
- llms.txt
- A plain-text guide to a site for AI tools. The scan looks for it but never scores it.
- MCP (Model Context Protocol)
- The open standard AI apps use to connect to external tools and data. An MCP server lists tools (
tools/list) and runs them (tools/call). - OpenAPI
- A standard format for describing an HTTP API. Transica builds connectors from OpenAPI specs.
- Own domain
- Hosting your connector at an address on your domain, such as
mcp.yourdomain.com. Business and up. - PKCE
- A security step in OAuth sign-in that directories require (with the S256 method).
- Slug
- A connector's short name, used in its address:
https://<slug>-mcp.transica.dev/mcp. - Streamable HTTP
- The transport remote MCP servers use. Transica hosts and connects to servers over it.
- Success check
- How a run is graded, independently of the agent: answer mentions, reaches page, page shows text.
- Task set
- A saved list of up to 10 tasks with their checks.
- Test account
- A dedicated login on your site that agents use for signed-in testing.
- Test credits
- The unit for AI-heavy work: 1 credit is one website run on the standard model.
- Test prompts
- Prompts that should (positive) or shouldn't (negative) make an assistant call your tools. Required by the directories.
- Tool
- One action an AI app can call on your connector, with a name, description, inputs and hints.
- Version
- A snapshot of a connector's tools: draft until published, kept forever once published.
- WebMCP
- A proposed browser standard that lets a web page offer tools to an AI agent in the browser. The scan shows it but doesn't score it.
- Write approval
- An admin's permission for a tool that changes data to run when hosted.