microsoft/autogen · warning · Error

Failed to get MCP prompt

Error message

Failed to get MCP prompt

What it means

Input-validation error from run() when args is not one of the three accepted shapes: a str, a dict containing a 'query' key, or a SearchQuery instance. Any other type (int, list, dict without 'query', None) is rejected immediately.

Source

Thrown at python/packages/autogen-studio/frontend/src/components/views/mcp/api.ts:177

  async getPrompt(
    serverParams: McpServerParams,
    name: string,
    promptArgs?: Record<string, any>
  ) {
    const response = await fetch(`${this.getBaseUrl()}/mcp/prompts/get`, {
      method: "POST",
      headers: this.getHeaders(),
      body: JSON.stringify({
        server_params: serverParams,
        name: name,
        arguments: promptArgs || {},
      }),
    });

    const data = await response.json();
    if (!response.ok) {
      throw new Error(data.message || "Failed to get MCP prompt");
    }

    return data;
  }

  async getCapabilities(
    serverParams: McpServerParams
  ): Promise<GetCapabilitiesResponse> {
    const response = await fetch(`${this.getBaseUrl()}/mcp/capabilities/get`, {
      method: "POST",
      headers: this.getHeaders(),
      body: JSON.stringify({ server_params: serverParams }),
    });

    const data = await response.json();
    if (!response.ok) {
      throw new Error(data.message || "Failed to get MCP capabilities");
    }

View on GitHub (pinned to 027ecf0a37)

Solutions

1. Pass a str (query text), {'query': '...'}, or a SearchQuery instance — matching the key name exactly.
2. If forwarding LLM arguments, validate/normalize them to {'query': ...} first.
3. For schema drift, consider parsing with SearchQuery.model_validate_json(args_json) to get pydantic's clearer error.

Example fix

# before
await tool.run({"search": "hotels in seattle"})  # wrong key -> ValueError

# after
await tool.run({"query": "hotels in seattle"})
# or
await tool.run(SearchQuery(query="hotels in seattle"))

from autogen_ext.tools.azure._ai_search import SearchQuery

def coerce_query(args):
    if isinstance(args, SearchQuery):
        return args
    if isinstance(args, str):
        return SearchQuery(query=args)
    if isinstance(args, dict) and isinstance(args.get("query"), str):
        return SearchQuery(query=args["query"])
    return None  # caller returns an error message to the agent

from typing import Any, Union

def is_valid_search_args(args: Any) -> bool:
    if isinstance(args, str):
        return True
    if isinstance(args, dict):
        return "query" in args
    return type(args).__name__ == "SearchQuery"

try:
    results = await tool.run(raw_args)
except ValueError as e:
    if "Invalid search query format" in str(e):
        return {"error": "expected str, {'query': ...}, or SearchQuery"}
    raise

When it happens

Trigger: Calling tool.run({'text': 'hotels'}) (wrong key), tool.run(['hotels']), tool.run(None), or tool.run(42). Common with hand-written tool-call plumbing that forwards raw JSON of the wrong schema.

Common situations: LLM function-call arguments that don't match the SearchQuery schema (e.g. key named 'search' or 'q'); passing args unpacked from a tuple; deserializers returning something other than the documented shapes.

Related errors

• Failed to list MCP prompts
• Failed to get MCP capabilities
• File {path} does not exist.
• expand_scenario expects an str or list for 'template'
• Tool names must be unique: {tool_names}

AI-assisted analysis of microsoft/autogen@027ecf0a37 (2026-08-15). Data as JSON: /api/errors/fc8feb1507e70137. Report an issue: GitHub.