feat: Add accept-list and reject-list repo filters

[akbxr]

Closes #247

This PR introduces new filtering capabilities to the CLI tool for repository selection and updates the documentation accordingly. It adds support for --accept-list and --reject-list options, enabling users to include or exclude repositories based on patterns. Additionally, it integrates these filters into the search results processing logic and enhances the user experience with detailed feedback about applied filters.

Before	After

[enesgules]

Thanks for your contribution. We’re still reviewing it and thinking about how we can further improve the developer experience of this feature. So it might take a bit of time before we can merge this PR.

[kylesnowschwartz]

I'd like to add my input that this is a highly desired feature that is currently preventing my team from adopting this tool. The fear is that we may inadvertently supply the AI with instructions to install or perform a destructive action (prompt injection attacks). By limiting what Context7 can look up with a defined list of approved repos we would limit that attack vector.

Thanks for your hard work @enesgules !

[enesgules]

We felt that a better solution was possible and we are working on it atm. Please let us know what you think @kylesnowschwartz

[kylesnowschwartz]

It's great that you're working on a robust solution, thank you. I'll add a few of my thoughts:

• In addition to a GUI, being able to pass these as arguments to the MCP server via the JSON config for a cli-first solution would be (more) convenient
• The list of documentation types from Github repos would ideally also be configurable for full flexibility
• 'Documentation Websites' remains pretty opaque. Maybe there's a way to make that more specific or discoverable what constitutes a documentation website?
• More verbose tool tips for full-text explanations of what adding a trusted org does (does it bypass the toggles above? Does it include all files or just the 'allowed-list' of file types? etc.). In fact just more verbose tool tips I think are best, i.e. "explain it like I'm 5" principle.

I'll share this around my team as well to see if there are any thoughts but this looks like great progress, thank you for engaging and supporting AI tooling! Cheers.

[enesgules]

this is now available through context7.com/dashboard

we have decided to move with this approach to keep the MCP and endpoints fairly simple. If we want to implement a feature such as this, we would have to either filter on the client side, which is slow and wouldn't work for every use case since the endpoint returns a limited number of results. or we have to send the filters to the server which would complicate the endpoints.

thanks for your patience and your contributions especially @akbxr <3