MCP

The STP.Documents MCP server makes document-related functionality available as tools to AI agents. It is available at https://common.stp-cloud.de/documents3/bot/mcp and supports stateless Streamable HTTP.

Authorization

The STP.Documents MCP server is an STP.Identity protected resource. Unfortunately STP.Identity does not support Client ID Metadata Documents or Dynamic Client Registration, so every MCP host using the STP.Documents MCP server needs an STP.Identity client registration (just like for the Universal API).

The Documents MCP server requires the license lexdmsmcp, license module documents3-mcp, permission documents3-mcp, scope doc3.mcp3 or role Documents3.McpUser. There is no default group for this role available, so it has to be created manually.

The doc3.mcp3 scope is intended to be used with spec compliant OAuth authorization, which requires the use of the MCP server’s resource indicator during token acquisition. That limits the token’s scopes to only scopes dedicated to the STP.Documents MCP server. When the STP.Documents MCP Server receives this sole scope, it uses token exhange internally to get another accesstoken to invoke downstream STP.Documents APIs on behalf of the user.

Example code can be found at STP.Documents.Example.

Tools

It provides the following tools:

stp_doc_get_current_user

Get information about the current user.

stp_doc_search_contexts

Find matters or other contexts. Loading happens via the Universal API.

stp_doc_list_documents

List documents of a matter (Akte) or other context. Loading happens via the Universal API.

stp_doc_search_documents

Find documents whose contents is semantically similar to the query string. This requires all findable documents to have been indexed. Loading happens via the Universal API.

stp_doc_get_document_metadata

Get document metadata like contexts, matters, or named values. This includes the same metadata available via the Universal API.

stp_doc_read_document

Get the textual content of the document. The fulltext rendition (stp.doc.fulltext.txt) is ready by default. If there is no fulltext rendition but the document ContentType is either text/plain, text/markdown, text/xml, application/xml, or application/json, the original rendition is read. If the original rendition is application/pdf, it extracts the text layer on-demand. If the original is any other type, it loads the preview rendition (which is always PDF) and extracts the text layer on-demand.

Loading happens via the Universal API.

stp_doc_create_document

Create a new document with text content (extension .txt). Access happens via the Universal API.

stp_doconprem_list_tasks

List all my open tasks (on-premise). Loading happens via unencrypted requests through Channel.OnPremise.

stp_doconprem_complete_task

Complete a single task or task list step (on-premise). Access happens via unencrypted requests through Channel.OnPremise.

stp_doconprem_search_containers

Find on-premises stored matters (Akten) or other containers. Loading happens via unencrypted requests through Channel.OnPremise.

stp_doconprem_list_documents

List on-premises stored documents of a matter (Akte) or other on-premise container. Loading happens via the Channel.OnPremise. Loading happens via unencrypted requests through Channel.OnPremise.

stp_doconprem_read_document

Get the textual content of the on-premises stored document. It always loads in PDF format and extracts the text layer on-demand. Loading happens via the Channel.OnPremise. Loading happens via unencrypted requests through Channel.OnPremise.

Resources

It provides the following resources:

stpdocuments://contexts/{contextReference}

Get the documents list of the context reference.

stpdocuments://documents/{documentId}/fulltext

Get the fulltext rendition of the document.

stpdocuments://documents/{documentId}/preview

Get the preview rendition of the document.

stpdocuments://documents/{documentId}/original

Get the original rendition of the document.