MediaKind Context MCP

Connect your AI tool to the MediaKind Model Context Protocol (MCP) server to find published documentation and Application Programming Interface (API) references. Search returns concise sources with links. You can then read a specific document or endpoint in more detail.

Connect your AI tool

Use this server address with a client that supports Streamable HTTP:

https://docs.mediakind.com/api/mcp

The server does not require an API key or authentication token. Its tools read documentation and do not modify your MK.IO resources.

For clients that accept a transport configuration, use:

{
  "transport": "streamable-http",
  "url": "https://docs.mediakind.com/api/mcp"
}

The client manages the MCP session. After connecting, call list_corpora to verify the connection and see the current documentation collections and API names.

Find the right source

Search the collection that matches your question. Use search_all_mkio_docs when the question spans multiple collections.

Tool	Use it to
list_corpora	See available collections, API names, and the catalog revision.
search_all_mkio_docs	Search MK.IO, MK.IO Beam, API guides, and API references together.
search_mkio_docs	Find MK.IO product guides and concepts.
search_beam_docs	Find MK.IO Beam configuration and operation documentation.
search_api_guide	Find API authentication instructions, workflows, and concepts.
search_api_reference	Search endpoint titles, tags, paths, and descriptions, optionally within one API.

The search tools require a query. For example, call search_mkio_docs with:

{
  "query": "How do I create a streaming locator?"
}

Product and guide searches return up to five distinct documents with short excerpts. Each result includes a document id, a title, and a canonical url. Results are JSON objects returned as text. A search with no relevant results reports no_matches.

Browse and inspect endpoints

Use catalog tools when you need an exact endpoint or a complete list of operations. These tools do not substitute a similar endpoint for a missing one.

Tool	Inputs and result
list_api_endpoints	Browse endpoints. Optional api_name and tag filters narrow the list.
search_api_by_tag	Browse endpoints with the exact resource tag, such as Assets.
get_api_endpoint	Match an HTTP method and exact path. Add api_name if the signature exists in more than one API.
compare_api_versions	Find endpoints matching a query, grouped by current API. This compares APIs, not historical versions.

Call list_api_endpoints to browse the Media API’s asset operations:

{
  "api_name": "media-api",
  "tag": "Assets",
  "limit": 20
}

The response includes total and next_cursor. If next_cursor is not null, pass it as cursor with the same filters to continue. Listing defaults to 20 entries and accepts a limit from one to 50. A response may contain fewer entries to stay within its size limit.

Use the API names returned by list_corpora. The aliases media, management, fleet, and infra are also supported. Unknown API names produce an input error rather than silently searching a different API.

An exact lookup reports not_found when no signature matches, or ambiguous when you need to select an API. A successful lookup returns the endpoint identity and document id for reading its details.

Read document details

Use get_document with an id returned by search or endpoint browsing. Replace the placeholder below with that value:

{
  "id": "<DOCUMENT_ID>"
}

The response includes document content and section headings. To read a particular section, pass its exact heading as section with the same id.

For an API endpoint, set format to openapi to read the exact source operation. References remain as $ref values. To follow a local reference, pass its value without the leading # as pointer, with the same endpoint id and format. Pass an empty pointer to read the complete specification, including shared parameters, security schemes, and server definitions.

OpenAPI content is JSON text. For a paginated response, concatenate the content fields from every page before parsing the JSON. Keep the same format and pointer when continuing. Markdown is a readable overview; use OpenAPI format when you need the complete schema.

Search responses are limited to 8 KiB. Document reads are limited to 16 KiB per response. Large documents return truncated: true and a next_cursor. Pass that value as cursor, with the same document and section, to continue. Truncated content is incomplete: read the continuation before relying on the full schema or procedure.

Handle updates and errors

Cursors belong to a catalog revision and a set of filters. If the documentation changes or you change filters, start again without the cursor.

The catalog refreshes within one minute. During publication, a search can report index_updating or incomplete while indexed content catches up with the current catalog. Retry shortly. A document that changes between discovery and reading reports content_changed rather than returning mismatched content.

Blank queries, queries containing only punctuation, and queries longer than 2,000 characters produce input errors. Service failures are reported as errors, separately from a valid search with no matches.