Can you add firecrawl for fetch markdown ?

[zx9597446]

Very cool project! I’ve tested many deep research tools, and this one is truly impressive.

However, the embedded browser seems quite unstable, and I couldn’t get Chrome remote debugging to work properly.

Additionally, r.jina.ai has rate limits, which makes it less suitable for local research.

I think Firecrawl could be a great solution for local research since it’s self-hosted and doesn’t rely on external rate limits.

[benhaotang]

Hi, just seeing this, I have been experimenting friecrawl both locally and using api and yes I think their performance is more stable.
But it may take some time for this release as I am writing my thesis recently. Will get back to improving this after 10th April.

[zx9597446]

Some suggestions for reference: consider directly configuring an MCP server to complete search and fetch markdown. There are already many such tools, so you can switch tools at any time. This project should focus on optimizing the deep research process.

[benhaotang]

Hi, I agree, and adding mcp is on my roadmap. But since this repo also covers some local models, and most local models cannot call tools reliably as a multiturn agent, in the end I will still hardcode some functions for them and add mcp support for online models.

[zx9597446]

can't wait looking forward to it

[zx9597446]

I believe local model will be more reliable to make tool calls, you should consider a clear ,simple plan.

[zx9597446]

and consider about expose as a mcp server too, since there is tool convert mcp server to open ai like endpoint