429 response

[niktor76]

Today, an MCP client sent multiple requests in a short period of time. One request succeeded, but the others failed with a 429 (Too Many Requests) response.

It would be helpful if the MCP server could handle this scenario by adding a small delay between multiple incoming requests within a short time window, preventing the 429 responses.

[jonathansampson]

Hello, @niktor76. It is typically the responsibility of the client to back-off when the server responds with a 429. Can you share a bit more information about the setup and scenario? Perhaps there is something I am not considering? Thank you in advance for any additional context/insight you can provide.

[niktor76]

HI @jonathansampson, thx for your response. It happened in VS Code Github Copilot Chat. I'll try to reproduce it. As I remember I asked the agent to search for different topics and it fired three requests to the brave search mcp at the same time.

[niktor76]

So I just reproduced it in Copilot chat using the Agent mode "Agent" and LLM "Raptor"
I hade to remove my workaround in my instruction file, which lets the agent wait 3 seconds before sending another request.

This model was smarter, and retried to request the failed requests sequentially:

[hspencer77]

@niktor76 , looking at the error, the Brave Search API plan associated to your deployment is Free tier. Per documentation - https://api-dashboard.search.brave.com/app/plans - the Free tier allows for the following:

• 1 request per second
• 2,000 requests per month

Given that you are doing three to five requests in parallel, this is why you are getting the 429. Are you suggesting (feature request) to add in a flag which introduces delays per each request?

[niktor76]

@niktor76 , looking at the error, the Brave Search API plan associated to your deployment is Free tier. Per documentation - https://api-dashboard.search.brave.com/app/plans - the Free tier allows for the following:

• 1 request per second
• 2,000 requests per month

Given that you are doing three to five requests in parallel, this is why you are getting the 429. Are you suggesting (feature request) to add in a flag which introduces delays per each request?

I see. Yes, I use the free tier. The flag would be actually a good idea.