Feature request for product/service
Cloud Agent (GitHub, Slack, Linear,…)
Describe the request
It would be helpful to have (in addition to the WebFetch tool) a built-in tool that can fetch content from the web safely. By safe, I mean it fetches the content, ensures the content won’t poison the model, then returns the safe content to our agent.
Why? In cloud environments, we intentionally restrict outbound network connections to avoid accidental poisoning. However, it’s common to issue a cloud agent prompt like “please modify this app so it works like this article/tweet https://example.com/123” Since example.com isn’t in our cloud env’s default allowlist (and it can be any arbitrary domain), there’s no way for the agent to fetch the full content.
We have worked around this by creating our own authenticated MCP Server Tool that uses Firecrawl and an LLM call to inspect the content before returning it to the user. Should just be baked into Cursor e.g. WebFetchSafe
Admittedly, this might not work for some bash tools or scripts to fetch remote content, but it’s a lot better than nothing.
Operating System (if it applies)
MacOS