🔍 y tho
I created a create-mcp-server based off of AWS well-architected patterns for MCPs (preventing tool sprawl etc) and wanted to test it out. Also, I couldn’t find an MCP server for hiring.cafe.
what
I still struggle with when is it sufficient to have LLMs just scrape the web browser with curl or when an MCP makes sense, but to try it out I created a simple HiringCafe MCP that has 4 simple tools. One of the factors I was trying to understand is whether providing more details in the tool is more efficient, or if giving it a summarized view with the option of fetching the specific one. I will be testing both.
search_jobs: search listings by the common filters, query/location into a set of normalized job resultsget_job: get a single jobs detailssearch:search_jobsbut a summarized view for ChatGPTfetch:get_jobbut summarized view
In addition to the main streamable-http MCP endpoint /api/mcp, there is also an /api/healthz endpoint that acts as a liveness probe.

