The `enhance` and `extract` tools always throw a server-side error when called, making them unusable via MCP.

[connor11528]

Bug

The enhance and extract tools always throw a server-side error when called, making them unusable via MCP.

Error

Error calling tool 'enhance': Invalid variable type: value should be str, int or float, got None of type <class 'NoneType'>
Error calling tool 'extract': Invalid variable type: value should be str, int or float, got None of type <class 'NoneType'>

Root cause

The tool schemas declare several parameters as anyOf: [string, null] (e.g. location, employer, title, school, page_type), meaning null is a valid value per the schema. However the Python server rejects None at runtime with the
above error.

This happens whether the optional fields are passed explicitly as null, passed as empty strings, or omitted entirely — the MCP framework appears to populate unset optional fields as None, which the server then rejects.

Reproduction

Call enhance with any valid Person lookup:

{                                                                                                                                                                                                                                                 
  "type": "Person",       
  "name": ["Connor Leech"],
  "email": ["connor@example.com"],
  "employer": "Acme",                                                                                                                                                                                                                             
  "title": "Engineer",
  "url": [],                                                                                                                                                                                                                                      
  "location": null,                                                                                                                                                                                                                               
  "school": null
}                                                                                                                                                                                                                                                 
                          
Call extract with just a URL:
{
  "url": "https://example.com",
  "format": "markdown"         
}                                                                                                                                                                                                                                                 
 
Both fail. search_web works correctly.                                                                                                                                                                                                            
                                                                                                                                                                                                                                                  
Expected behavior
                                                                                                                                                                                                                                                  
Optional parameters that are null or omitted should be handled gracefully by the server (ignored or treated as unset).                                                                                                                            
 
Environment                                                                                                                                                                                                                                       
                          
- Remote MCP server: https://mcp.diffbot.com/mcp?token=...                                                                                                                                                                                        
- MCP client: Claude Code CLI

[kandiou]

Investigated this — the root cause isn't the null optional parameters. It's the token never being resolved, which leaves params["token"] = None, and aiohttp/yarl refuses to build a query string containing None. The wording in the error is aiohttp's own.

params = {
    "token": diffbot.token,   # <- None
    "url": url,
}

Evidence:

1. The same error occurs with a deliberately invalid token, and with extract {"url": "https://example.com"} alone (no optional params at all). So it fails before authentication and independently of the arguments — which rules out the anyOf: [string, null] schema theory.
2. search_web is unaffected because it passes the token in a header built with an f-string: f'Bearer {diffbot.token}'. None becomes the literal string "None", so no exception is raised. That's the only reason it looks healthy.
3. Running this exact diffbot_mcp_server.py locally with DIFFBOT_TOKEN set in the environment, extract and enhance both work, with no code change. A second rapid call returns a normal 429 Too Many Requests, confirming the token is genuinely being sent.

So the server code isn't broken. The hosted deployment at mcp.diffbot.com appears to have no DIFFBOT_TOKEN in its environment, and the per-request token resolution in DiffbotAPI.__init__ isn't taking effect there either — hence None for every caller, regardless of the token they supply.

Suggested hardening, independent of the deployment fix: raise a clear error when no token resolves, instead of letting aiohttp surface a confusing type error.

if not self.token:
    raise ValueError("No Diffbot token: set DIFFBOT_TOKEN, or pass ?token= / Authorization: Bearer")

Workaround for anyone blocked: self-host this repo with DIFFBOT_TOKEN in the environment. All three tools work.