Improve tool description for ask_perplexity

[digitarald]

The current description doesn't say anything about when and how to use the tool (see https://docs.anthropic.com/en/docs/build-with-claude/tool-use/overview#best-practices-for-tool-definitions).

https://github.com/ppl-ai/modelcontextprotocol/blob/2971e038dc05ff94202dc5fae92ca423cb42a11c/perplexity-ask/index.ts#L19-L21

To further improve reliability, I suggest simplifying the parameters to just the research prompt and not a user/assistant message array.

[minovap]

I wanted to change or at least fine tune most of the MCP tool descriptions i encountered for my own use cases, so i created this MCP proxy server that can override the description for the tools. It can also hide/disable tools from MCP servers so Claude's context stays as clean possible with only the tools I want visible. However the MCP server is built around batching MCP tool requests, so it is not a pure proxy.

https://github.com/minovap/mcp-gateway

[digitarald]

Related, we have an issue open for VS Code to allow customizating tools/servers. I'd love to see comments there on specific areas where devs needed it: microsoft/vscode#243940