uip docsai

uip docsai asks natural-language questions against UiPath's documentation, returning an AI-generated answer with source URLs and suggested follow-up questions. Internally, the tool talks to a hosted MCP server (SearchDocumentation tool) rather than a REST endpoint — this is why the output shape has a query_id field.

See Tools (plugins) for how this tool integrates with the CLI.

Synopsis

uip docsai ask <query> [-t <tenant>] [--source <docs|technical_solution_articles>]

All subcommands honor the global options and exit codes.

uip docsai ask

Search UiPath documentation. Returns a synthesized answer and the sources that backed it.

Arguments:

• <query> (required) — The question, as a natural-language string. Wrap in quotes if it contains spaces.

Options:

• -t, --tenant <name> — Tenant; defaults to the tenant selected by uip login.
• --source <source> — Knowledge base to search. One of docs (default) or technical_solution_articles. Invalid values fail at parse time.

Examples:

# Ask a how-to question against the main docs
uip docsai ask "How do I create a process?"

# Search the technical solution articles instead
uip docsai ask "Selector failed with 404" --source technical_solution_articles

Output formatting

uip docsai ask adapts its output to the CLI's global --output mode:

• --output table (default) and --output plain print a readable, human-friendly response: the answer text, then a Sources: list, then (if present) Follow-up questions:.
• --output json emits the parsed response in the standard envelope.

Data shape (--output json)

{
  "Code": "DocsAIAsk",
  "Data": {
    "query_id": "a1b2c3d4-0000-0000-0000-000000000001",
    "text": "To create a process, open UiPath Orchestrator, navigate to Processes, and click Add.",
    "sources": [
      "https://docs.uipath.com/orchestrator/standalone/2022.4/user-guide/managing-processes"
    ],
    "disclaimers": [],
    "followups": [
      "What permissions are required to create a process in Orchestrator?"
    ]
  }
}

• query_id — unique ID for this query.
• text — AI-generated answer.
• sources — absolute URLs for the documentation pages cited in the answer.
• disclaimers — any cautions the server wants the client to surface alongside the answer (often empty).
• followups — suggested next questions, ranked.

If the server returns a response the CLI cannot parse as the structured shape above, Data falls back to the raw content array from the MCP tool call so nothing is lost.

Error behavior

• Failure to create the MCP client (typically an auth or connectivity issue): Failure: "Error querying DocsAI" with the underlying error in Instructions, exit 1.
• MCP tool call errors: same Failure: "Error querying DocsAI" envelope; the CLI closes the client cleanly before exiting.
• Server-side error (e.g., rate limit): Failure: "DocsAI returned an error" with the raw content payload in Instructions.

Related

• Tools (plugins) — install, update, uninstall the docsai-tool package.
• uip mcp — lets a client MCP host invoke the CLI; docsai goes the other direction and calls a remote MCP server.
• uip login — session used to authenticate DocsAI calls.