MCP tool search timeout with Amplitude toolkit We're integrating Composio via the MCP server URL (composio.create() → session.mcp.url) using @composio/core@0.6.7 with @mastra/mcp. When a user connects Amplitude (API_KEY auth, 54 tools), calling COMPOSIO_SEARCH_TOOLS consistently times out at 30s: MCP error -32001: Request timed out timeout: 30000 The session creates fine, listTools() works, but any runtime COMPOSIO_SEARCH_TOOLS call from the agent hangs until timeout. This doesn't happen with smaller toolkits. Setup: - @composio/core: 0.6.7 - MCP client: @mastra/mcp (MCP SDK @modelcontextprotocol/sdk@1.26.0) - Transport: Streamable HTTP via session.mcp.url - Timeout: 30s (default) Question: Is this a known latency issue with COMPOSIO_SEARCH_TOOLS for larger toolkits over MCP? Is there a recommended timeout or a way to warm up the search index per session?

Hey @Narkon, Ideally, the

COMPOSIO_SEARCH_TOOLS

tool shouldn't timeout based on the no.of tools in a toolkit. Could you share a reproducible code snippet? Thanks!

Frankly after further investigation over next days it seems to be some infrastructure/network problem. Most of the time Agent can talk to your MCP server (url generated by creating a session), but from time to time it's getting stuck and timeouts (30s timeout). Could you please check logs on your side are you getting our requests? Example search tool call that failed:

{
  "queries": [
    {
      "use_case": "send message to channel using slackbot toolkit"
    }
  ],
  "session": {
    "id": "wave"
  }
}

Hey @Narkon, thanks for information, we are checking this.

I'm seeing this as well. Was there ever a resolution @Uday?

Hi @Cashed yes, we've done load testing and things seemed to be working fine. That said, there can be instances where you might encounter this. Could you let me know how often you're running into it? Any log IDs would also be super useful. Thanks!