Dataverse People Lookup

Dataverse People Lookup

Answer people-related queries (title, location, manager) by driving the four-step Dataverse MCP pipeline end-to-end.

When to Use

• "What is [name]'s title, location, and manager?"
• "Who does [name] report to?"
• "Where is [name] based?"
• "What role does [name] have?"
• Any combination of the above for one or more people.

Workflow

Use the CallMcpTool tool with server: "user-DataverseMCP" for every step.

Step 1 — Identify the data product

toolName: identify_dataproducts
arguments: { "user_query": "<the user's question verbatim>" }

Inspect the returned list. Pick the first (most relevant) data product name for the next step. If more than one data product is flagged (cross-domain), follow the ordering instructions in the response.

Step 2 — Shortlist tables

toolName: shortlist_tables
arguments: {
  "data_product": "<data product from Step 1>",
  "user_query": "<the user's question verbatim>"
}

The response is a list of table objects with name, schema, and description. Save the full list for Step 3.

Step 3 — Generate SQL

toolName: get_sql
arguments: {
  "data_product": "<data product from Step 1>",
  "tables_list": <table list from Step 2>,
  "user_query": "<the user's question verbatim>"
}

The response contains a sql field. Extract the SQL string for Step 4.

Step 4 — Execute SQL

toolName: execute_sql
arguments: { "sql": "<SQL from Step 3>" }

The response contains columns, data, and row_count.

Presenting Results

After Step 4, format the answer for the user:

• Single person — present as a short summary:

Jane Doe

• Title: Senior Data Engineer
• Location: London
• Manager: John Smith

• Multiple people — use a bullet list with the same structure per person.

• If any field is null or missing, say "Not available" instead of omitting it.

Error Handling

• If Step 1 returns no data products, tell the user the query could not be mapped to a known data domain and ask them to rephrase.
• If Step 4 returns row_count: 0, tell the user no matching records were found and suggest checking the spelling of the person's name.
• If any step fails with an error, surface the error message to the user and suggest retrying or refining the query.

Important Notes

• Always pass the user's original question as user_query — do not paraphrase or simplify it, because the downstream LLM uses it for context.
• Do not skip steps or call execute_sql directly for people lookups; the full pipeline ensures correct table selection and business-rule compliance.
• For follow-up questions ("What about their email?"), restart from Step 1 to ensure proper context is applied.