Select the statement that best describes the organization today, one per
chapter. The result shows a level for each chapter and the chapters to read
first. Answers are stored in this browser only.
1. Purpose How is the scope of agent access defined?Not yet It is not; agents reach whatever the website exposes. Foundational The organization has decided which data and which uses agent access is for, and that only public data are in scope. AI-ready Use cases, scope, and success criteria are written, with a question set that represents them. AI-native The scope is reviewed from the evaluation results and usage, and the question set grows with real questions. 2. Tools How are the tools designed?Not yet There is no agent interface; agents scrape pages or call the API ad hoc. Foundational The API is documented with OpenAPI or SDMX and agents can call it. AI-ready An MCP server offers a small set of read-only tools with documented inputs, structured outputs, and guidance resources, checked against design rules. AI-native Tools are versioned, their descriptions are tested with models, and the manifest is published. 3. Context What context does the server give the assistant beyond the tools?Not yet None; the tool descriptions are all there is. Foundational A guidance resource states what the server provides, how to cite, the licence, the limitations, and when to refuse; code lists are resources. AI-ready Calendar and per-series notes are resources with templates and annotations; prompts cover the common question types; the guidance is checked against the manifest. AI-native The effect of each guidance change is measured with the question set, and the guidance is versioned with the server. 4. Provenance What does a data response carry?Not yet Values only. Foundational Values with unit and period. AI-ready Series identifier, unit, observation status, release date, source URL, licence, and citation in every response. AI-native The same, with version identifiers and freshness, and answers are verified against the responses. 5. Access Who may call the interface, and how much?Not yet Anyone, without limits. Foundational Public access with a per-client rate limit; terms of use published. AI-ready Tiers with keys, quotas, accepted terms, and contacts in a registry; limits enforced; the log audited against the registry. AI-native OAuth-based authorization for user-level access with scopes per tool group; usage per client reported to the monitoring. 6. Safety How is the interface kept safe?Not yet No specific measures. Foundational Only public data are exposed; the server is read-only. AI-ready Access control, rate limits, input validation, output sanitization, and logging are in place; prompt-injection tests run. AI-native Safety checks run on every change, the server is in the AI component register, and incidents are reviewed. 7. Evaluate How is agent access evaluated?Not yet It is not. Foundational A few questions are tried by hand after changes. AI-ready A versioned question set with expected tools, values, and behaviour is run through an agent, and the traces are scored. AI-native The suite runs on every change, per language, with a reviewed sample, and results are kept with the version. 8. Distribute How do users and developers find and connect to the server?Not yet By asking the team. Foundational A developer page gives the endpoint, one configuration, the terms, and a contact. AI-ready Configurations for the common assistants are published and tested each release; the server is listed in the registry with a checked listing. AI-native A package for local use is published beside the endpoint, and the listing and configurations update from the manifest and version. 9. Operate How is the server operated?Not yet A prototype on a laptop. Foundational A hosted server with a stable endpoint and a contact. AI-ready Versioned, monitored, with published client configuration, usage statistics, and a cost estimate. AI-native Usage feeds monitoring and documentation work; deprecations are announced; the server is listed for clients. 0 of 9 answered Show reading path Reset
Interpretation of the result
Chapters 1, 5, and 6 come first: scope, access, and safety are
decisions, and they are cheap compared with their cost when skipped.
Chapter 2 depends on the API and catalog described in the dissemination
cookbook; a server over an undocumented API inherits every gap, and
chapter 3 is where the gaps the tools cannot close are stated.
Chapter 7 is the gate for publication: a server without an evaluation
suite cannot be changed safely.
Chapter 8 is the one most often forgotten; a server nobody can configure
has no users.