Claude 3.7 Sonnet causes Actions section to be greyed out as per image.
1 Like
Hi @ecomprocess, this is because 3.7 is a reasoning model, and actions do not work too well with them. They have greyed out the action tab for all the reasoning models like o1, o3 and Gemini 2.5 to avoid issues with your pickaxes.
1 Like
Hi @ecomprocess as a general rule, Connected actions function much better with OpenAI models - with the exception of MCP servers and prompt chained Pickaxes that tend to work well with many (if not all models).
1 Like
@Ned.Malki @ab2308 Which model do you think would connect best with MCP based on your experience? Thank you.
2 Likes
Hi @chipvn That’s a great question Based on my experience, the simple answer is that you can’t go wrong with the most popular, up-to-date models.
Think of MCP as a universal translator. It’s designed to let any modern AI model “talk” to external tools. So, the good news is that they all connect perfectly fine.
The real difference is how smart the AI is. For the best, most reliable results, I recommend using a powerful, mainstream model like OpenAI’s GPT-4.1 (or mini) or X AI’s Grok4. They are particularly good at understanding what you want and figuring out how to use tools to get it done.
For most projects, the default model in Pickaxe,
GPT-4.1 mini, is a smart, fast, and affordable choice that works great.
1 Like
Hey @chipvn, the connection to the MCP server is established by the MCP client (in our case Pickaxe). Once connected, the server exposes the available tools/functions to the LLM. The LLM can then identify the tools (often through metadata) and decide when to use them.
While all the LLMs available in Pickaxe can work with tools (or function calling), some are better than others at interpreting metadata and using function calling. Claude is a good choice as the company behind these models (Anthropic) is the one that created the MCP standard. Open AI’s latest models are also working quite well.
I haven’t tested the other models with MCPs.
Hope it makes sense
1 Like