Strict validation fails for optional Tool/Prompt fields (title, icons, outputSchema, annotations)

[vavasilva]

Description

MCP Inspector fails with validation errors when connecting to a valid MCP server that doesn't include optional fields in Tool and Prompt definitions.

Error Messages

When clicking on Tools tab:

[
  { "expected": "string", "code": "invalid_type", "path": ["tools", 0, "title"], "message": "Invalid input" },
  { "expected": "array", "code": "invalid_type", "path": ["tools", 0, "icons"], "message": "Invalid input" },
  { "expected": "object", "code": "invalid_type", "path": ["tools", 0, "outputSchema"], "message": "Invalid input" },
  { "expected": "object", "code": "invalid_type", "path": ["tools", 0, "annotations"], "message": "Invalid input" },
  { "expected": "object", "code": "invalid_type", "path": ["tools", 0, "execution"], "message": "Invalid input" }
]

When clicking on Prompts tab:

[
  { "expected": "string", "code": "invalid_type", "path": ["prompts", 0, "title"], "message": "Invalid input" },
  { "expected": "array", "code": "invalid_type", "path": ["prompts", 0, "icons"], "message": "Invalid input" }
]

Expected Behavior

According to the MCP specification, these fields are optional:

• title (optional string)
• icons (optional array)
• outputSchema (optional object)
• annotations (optional object)
• execution (optional object)

The Inspector should not fail validation when these optional fields are not present.

Environment

• MCP Inspector: latest via npx @modelcontextprotocol/inspector
• Transport: STDIO via mcp-remote
• MCP Server: Python with mcp>=1.0.0
• OS: macOS

Steps to Reproduce

1. Create an MCP server with tools/prompts that only include required fields (name, description, inputSchema)
2. Connect via Inspector using STDIO transport with mcp-remote
3. Click on Tools or Prompts tab
4. See validation errors

Workaround

The server works correctly with Claude Desktop and Claude.ai. Only the Inspector UI shows these validation errors.

[vavasilva]

Workaround Found

We resolved the issue on our server side by adding the optional fields with default values:

For tools/list:

tool_dict["title"] = tool_dict.get("name", "")
tool_dict["icons"] = []
tool_dict["annotations"] = {}
tool_dict["execution"] = {}
# Remove outputSchema if None (we return TextContent, not structured content)
if tool_dict.get("outputSchema") is None:
    del tool_dict["outputSchema"]

For prompts/list:

prompt_dict["title"] = prompt_dict.get("name", "")
prompt_dict["icons"] = []

For resources/list:

res_dict["title"] = res_dict.get("name", "")
res_dict["icons"] = []
res_dict["annotations"] = {}

This workaround makes our server compatible with the Inspector's stricter validation.

However, the underlying issue remains: the Inspector should not fail validation for optional fields according to the MCP spec. Consider making the validation more lenient for optional fields.