How to Debug MCP Tool Calls

Blind agents and broken schemas

A common hurdle when building Model Context Protocol (MCP) integrations is the language model inventing arguments or hallucinating endpoints. Without visibility into the raw tool_use blocks, debugging schema mismatches or rate limit failures is a guessing game.

Disclaimer: wmcp.sh is an independent infrastructure provider and is not affiliated with Anthropic, OpenAI, or Microsoft.

Inspecting tool_use schemas dynamically

# 1. Fetch the exact tool schema the LLM sees
curl -s "https://wmcp.sh/api/v1/tools?url=https://acme-corp.example/openapi.yaml" | jq

# 2. Example output highlighting expected arguments
{
  "tools": [
    {
      "name": "get_customer_data",
      "description": "Retrieve customer information",
      "input_schema": {
        "type": "object",
        "properties": {
          "customerId": { "type": "string" }
        },
        "required": ["customerId"]
      }
    }
  ]
}

# 3. If the LLM sends "customer_id" instead of "customerId", the call will fail.
#    Use wmcp.sh verbose logging to catch this mismatch instantly.

Debugging Features

Common questions.