Tool reference

The server exposes five tools. All of them are deterministic reads that run no model, and all carry the same annotations:

Which tool for which taskCopy Anchor Link

Not available over MCP: courses, a module's scenes or script, building or publishing, and learner progress. Do those in the Vidext app.

ConventionsCopy Anchor Link

• Arguments are strict. Each tool accepts only the arguments listed here. An unknown tool, a missing required argument, an extra argument, or a value outside its limits is refused with JSON-RPC InvalidParams (-32602). Fix the call; retrying it unchanged cannot succeed.
• Organization selector. Every tool also accepts organizationSlug and organizationId (strings, 1–255 characters). Leave both out to use the organization bound to the token. See organization scoping.
• Results. A successful call returns its JSON object twice: as text in content[0].text and as structuredContent. A failure returns isError: true with an error object, except for the two knowledge tools, which report most failures in the result itself with success: false. See Errors and limits.
• IDs. moduleId is a UUID, as returned by list_modules.

search_knowledgeCopy Anchor Link

Search the organization's knowledge and return ranked excerpts, each with an opaque handle for read_knowledge_source.

It searches:

• the organization profile (org_profile);
• the source catalog, the title and metadata of every source the organization has added (source_registry);
• the processed text of those sources (source_content);
• a legacy organization wiki (generated_org_wiki), where an organization still has one. It is read but no longer updated.

Sources are added in the app, and a source is searchable once it has been processed.

Result. success: true, mode: "search", status ("indexed", or "empty" when there was nothing to search), the query, matchedResults, and results, ordered by relevance. Each result carries:

The result also carries diagnostic counters (such as retrievalTrace). They describe how the search ran; do not build on them.

In-band failures. success: false with status: "invalid" (an empty or too vague query) or status: "unavailable" (knowledge could not be read; reason: "timeout" when the 15-second budget ran out). Unavailable searches can be retried.

read_knowledge_sourceCopy Anchor Link

Read a bounded slice of one source, by a handle that search_knowledge returned. Arbitrary paths, and stale or altered handles, are refused, and organization access is checked again.

Result. success: true, mode: "read", the handle, sourceKind, title, path, documentId, sourceMetadata, and the slice:

In-band failures. success: false with status:

• "rejected": the handle did not come from search_knowledge for this organization, or no longer matches the source. Search again.
• "missing" or "empty": the source, or its processed text, is no longer there.
• "unavailable": knowledge could not be read right now (reason: "timeout" after 15 seconds). Retry.
• "invalid": no handle was given.

list_modulesCopy Anchor Link

List the organization's modules, most recently updated first. Admins and members see every module in the organization.

Result.

{
  "organizationId": "…",
  "modules": [
    {
      "id": "018f1111-2222-7333-8444-555555555555",
      "name": "Incident response",
      "description": "…",
      "color": "…",
      "published": true,
      "createdAt": "…",
      "updatedAt": "…"
    }
  ]
}

published is true while the module is live: its course is published and learners can see it. A module that is built but not live, or still a draft, has published: false. The result has no cursor; it returns at most limit modules, so narrow with search to find older ones. It does not include the module's course.

get_module_status (legacy)Copy Anchor Link

Read the outline of a module's planning run, from the earlier planning flow. Modules built in the chat by Vidext's agents have no planning run, so this tool has nothing useful to say about them. It never reports whether a module is built or live; use list_modules for that.

Result. organizationId, moduleId, and outline:

get_operation (legacy)Copy Anchor Link

Poll one outline run of a module until it ends.

Result.

{
  "operationId": "…",
  "moduleId": "018f1111-2222-7333-8444-555555555555",
  "kind": "outline",
  "status": "outline",
  "terminal": false,
  "pollAfterMs": 1500,
  "snapshot": { "…": "the outline object from get_module_status" }
}

• status is "waiting_for_approval" when the outline is ready, "failed" when it could not be written, and otherwise the run's current phase.
• terminal is true for waiting_for_approval and failed.
• pollAfterMs is present while the run is not terminal. Wait at least that long, then call again with the same arguments.

An operationId that is not the module's current run is refused with the tool error operation_not_found.