A connector establishes the connection. A connector tool exposes the functions. A Tool node is where your workflow calls a selected function and passes the inputs it needs.

This is the point where a user’s request becomes a specific operation, such as retrieving policy passages or searching a document library.

Start with a small workflow

A document finder can use a simple flow:

Start → Find SharePoint Files (Tool) → Show Search Results (LLM)

The Tool node calls sharepoint_search_items. It uses the approved library’s driveId and receives the search term from the current chat message. The LLM node then presents the filenames and links returned by the tool.

Workflow step What happens
Start Receives the user’s message.
Tool node Calls the selected connector function with the configured inputs.
LLM node Uses the returned information to prepare the response.

An indexed policy assistant can use the same workflow shape. Its Tool node retrieves relevant Content Intelligence passages, which the LLM uses to answer the question.

Map the inputs to the task

Select the connector tool and function, then configure the parameters shown in the function’s schema. Some values can stay fixed for the example, such as an approved library ID. Others should change with each request, such as the user’s question or search term.

Pass the Tool node’s output into the downstream LLM prompt using the actual node code. This gives the LLM the evidence needed for its response.

If the workflow needs another operation—such as retrieving metadata after the user selects a search result—add that call and provide the selected item’s identifiers.

Look at the output before tuning the prompt

Run a request with a known result and inspect the Tool node’s output. Then compare it with the final response.

Document passages can support an answer about a policy. File metadata can support names, links, and property checks. Keep the response aligned with the information returned, and test how the workflow handles an empty result or a failed tool call.

Once this small flow works, use the practical examples in the introduction to build on it.

Workflows, prompts, and results in the practical examples

The screenshots below are retained from the existing practical workflow guide. Their original figure numbers, captions, and image sizes are preserved.

WebCrawler workflow with START, Retrieve Q and A Tool node, and Answer User Query LLM node.
Figure 5. The WebCrawler workflow retrieves indexed web content and passes it to an LLM for a grounded answer.
WebCrawler workflow with the Answer User Query LLM node and its System Prompt and User Prompt displayed.
Figure 6. Configure the WebCrawler LLM node to answer only from the indexed web content returned by RETRIEVE_Q_A.
SharePoint workflow with START, Retrieve Chunks Tool node, and Answer User Query LLM node.
Figure 10. The SharePoint workflow retrieves indexed document chunks and passes them to an LLM for a grounded answer.
SharePoint workflow with the Answer User Query LLM node and its System Prompt and User Prompt displayed.
Figure 11. Configure the SharePoint LLM node to answer from the indexed content returned by RETRIEVE_CHUNKS.
AICOE DEMO Microsoft SharePoint File Finder with Start, Find SharePoint Files Tool, and Show Search Results LLM.
Figure 14. The simple search workflow uses the same Tool-to-LLM pattern as the earlier examples, but returns live file metadata rather than indexed document chunks.
Show Search Results LLM node with the simplified system prompt and dynamic Tool-output expression.
Figure 15. Configure the LLM with the simple prompt below. The user prompt receives the search output through an expression, not a hardcoded answer.
Successful SharePoint search workflow showing document links in chat and successful Tool and LLM nodes.
Figure 16. The search workflow returns matching document links using the user’s chat input. It does not summarize the files’ contents.

Back to Introduction to Connectors

Explore the practical connector workflows.