Bug: Invalid schema error when trying to use Context7 MCP: Overly strict schema validation? #4416
Description
System Info
Running latest Llama Stack on main right now
Information
- The official example scripts
- My own modified scripts
🐛 Describe the bug
I am trying to use the Context7 MCP server as shown in the snippet below. It fails with "Invalid schema for function 'search-library-docs': In context=('properties', 'folders'), array schema missing items". My guess is that this is correct that the schema is invalid. However, this is a very popular MCP server and other MCP clients like Cursor are able to handle it. So I think the real issue here is that we should be less strict with our enforcement of schema requirements.
from llama_stack_client import LlamaStackClient
client = LlamaStackClient(base_url="http://localhost:8321")
response = client.responses.create(
model="openai/gpt-4o",
input="Can I get any info about https://github.com/llamastack/llama-stack from Context7?",
tools=[{"type": "mcp", "server_url": "https://context7.liam.sh/mcp", "require_approval": "never", "server_label": "Context7"}]
)Error logs
Client:
---------------------------------------------------------------------------
InternalServerError Traceback (most recent call last)
Cell In[14], line 5
1 from llama_stack_client import LlamaStackClient
3 client = LlamaStackClient(base_url="http://localhost:8321/")
----> 5 response = client.responses.create(
6 model="openai/gpt-4o",
7 input="Can I get any info about https://github.com/llamastack/llama-stack from Context7?",
8 tools=[{"type": "mcp", "server_url": "https://context7.liam.sh/mcp", "requires_approval": False, "server_label": "Context7"}]
9 )
File ~/sample-agent/venv-3_12_9/lib/python3.12/site-packages/llama_stack_client/_utils/_utils.py:288, in required_args.<locals>.inner.<locals>.wrapper(*args, **kwargs)
286 msg = f"Missing required argument: {quote(missing[0])}"
287 raise TypeError(msg)
--> 288 return func(*args, **kwargs)
File ~/sample-agent/venv-3_12_9/lib/python3.12/site-packages/llama_stack_client/resources/responses/responses.py:248, in ResponsesResource.create(self, input, model, conversation, include, instructions, max_infer_iters, max_tool_calls, parallel_tool_calls, previous_response_id, prompt, store, stream, temperature, text, tools, extra_headers, extra_query, extra_body, timeout)
217 @required_args(["input", "model"], ["input", "model", "stream"])
218 def create(
219 self,
(...) 246 timeout: float | httpx.Timeout | None | NotGiven = not_given,
247 ) -> ResponseObject | Stream[ResponseObjectStream]:
--> 248 return self._post(
249 "/v1/responses",
250 body=maybe_transform(
...
-> 1054 raise self._make_status_error_from_response(err.response) from None
1056 break
1058 assert response is not None, "could not resolve response (should never happen)"
InternalServerError: Error code: 500 - {'detail': 'Internal server error: An unexpected error occurred.'}
Server:
INFO 2025-12-19 13:22:07,312 uvicorn.access:473 uncategorized: ::1:59762 - "POST /v1/responses HTTP/1.1" 200
ERROR 2025-12-19 13:27:50,200 llama_stack.core.server.server:281 core::server: Error executing endpoint
route='/v1/responses' method='post'
╭───────────────────────────────────── Traceback (most recent call last) ─────────────────────────────────────╮
│ /Users/bmurdock/sample-agent/venv-3_12_9/lib/python3.12/site-packages/llama_stack/core/server/server.py:271 │
│ in route_handler │
│ │
│ 268 │ │ │ │ │ return StreamingResponse(gen, media_type="text/event-stream") │
│ 269 │ │ │ │ else: │
│ 270 │ │ │ │ │ value = func(**kwargs) │
│ ❱ 271 │ │ │ │ │ result = await maybe_await(value) │
│ 272 │ │ │ │ │ if isinstance(result, PaginatedResponse) and result.url is None: │
│ 273 │ │ │ │ │ │ result.url = route │
│ 274 │
│ │
│ /Users/bmurdock/sample-agent/venv-3_12_9/lib/python3.12/site-packages/llama_stack/core/server/server.py:193 │
│ in maybe_await │
│ │
│ 190 │
│ 191 async def maybe_await(value): │
│ 192 │ if inspect.iscoroutine(value): │
│ ❱ 193 │ │ return await value │
│ 194 │ return value │
│ 195 │
│ 196 │
│ │
│ /Users/bmurdock/sample-agent/venv-3_12_9/lib/python3.12/site-packages/llama_stack/providers/inline/agents/m │
│ eta_reference/agents.py:115 in create_openai_response │
│ │
│ 112 │ │ metadata: dict[str, str] | None = None, │
│ 113 │ ) -> OpenAIResponseObject: │
│ 114 │ │ assert self.openai_responses_impl is not None, "OpenAI responses not │
│ initialized" │
│ ❱ 115 │ │ result = await self.openai_responses_impl.create_openai_response( │
│ 116 │ │ │ input, │
│ 117 │ │ │ model, │
│ 118 │ │ │ prompt, │
│ │
│ /Users/bmurdock/sample-agent/venv-3_12_9/lib/python3.12/site-packages/llama_stack/providers/inline/agents/m │
│ eta_reference/responses/openai_responses.py:425 in create_openai_response │
│ │
│ 422 │ │ │ │ │ if failed_response and failed_response.error │
│ 423 │ │ │ │ │ else "Response stream failed without error details" │
│ 424 │ │ │ │ ) │
│ ❱ 425 │ │ │ │ raise RuntimeError(f"OpenAI response failed: {error_message}") │
│ 426 │ │ │ │
│ 427 │ │ │ if final_response is None: │
│ 428 │ │ │ │ raise ValueError("The response stream never reached a terminal state") │
╰─────────────────────────────────────────────────────────────────────────────────────────────────────────────╯
RuntimeError: OpenAI response failed: Error code: 400 - {'error': {'message': "Invalid schema for function
'search-library-docs': In context=('properties', 'folders'), array schema missing items.", 'type':
'invalid_request_error', 'param': 'tools[1].function.parameters', 'code': 'invalid_function_parameters'}}
Expected behavior
I would like the MCP tool to just work like it does with Cursor. With that said, OpenAI's own implementation of the Responses API fails on this too with a similar error:
BadRequestError: Error code: 400 - {'error': {'message': 'Invalid tool schema for: Context7.search-library-docs', 'type': 'invalid_request_error', 'param': 'tools', 'code': None}}
So I guess it's debatable whether we want to fix this. FWIW, I'd prefer to be more robust than OpenAI in this case.