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 a client that supports Streamable HTTP. Choose the address for the documentation you need.
| Access | Server address | Documentation |
|---|---|---|
| Public | https://docs.mediakind.com/api/mcp |
Public MK.IO and MK.IO Beam documentation and API references. No sign-in required. |
| Signed in | https://docs.mediakind.com/api/mcp/private |
Public documentation and private product guides and API references your documentation account can read. Contributor guides are excluded. |
For signed-in access, use a client that supports OAuth. Open the authorization link from the client, review the requested access, and select Allow and sign in. Use your existing MediaKind employee or customer documentation account. Your client can refresh access while the documentation session is active. When the session expires, reconnect if your client prompts you.
Both addresses provide read-only tools. They do not modify your product 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 products, content kinds, and API names you can search.
Find the right source
Search the product that matches your question. Use search_all_mkio_docs when the question spans MK.IO and MK.IO Beam, or search_docs for any other product.
| Tool | Use it to |
|---|---|
list_corpora |
See available products, content kinds, API names, and the catalog revision. |
search_docs |
Search the documentation available through your connection. Optional product and kind (docs, api-guide, api-reference, or api-source) narrow the search. Use list_corpora to find valid products. |
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. Each result includes a document id, a title, and a canonical url. A catalog_lexical result matches the current title or path. Use its ID with get_document to read the source. 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 matching operations in current specifications, grouped by API name. |
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 or standalone bold section label as section with the same id. Clients can read structuredContent or parse the same JSON from the tool's text response.
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.
Large documents return truncated: true and a next_cursor. Pass that value as cursor with the same inputs to continue. Read all pages before relying on a complete 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 and exact document reads use the currently published documentation. The search index can lag behind a release. A signed-in search with incomplete results includes a reason and guidance. Try narrowing by product or content kind, or use get_document for a cited result. If a read reports content_changed, find the document again before retrying.
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.