Building Goatforce

Goatforce is a CRM dashboard powered by MCP servers, designed to be the one-stop shop for sales reps to manage their deals efficiently. The core idea was to eliminate the friction between different tools and create a unified interface that understands natural language commands. Goatforce has a sales reps Slack & Gmail logs and interactions with both all in one platform

The challenge was building a lightweight interface that could handle complex sales workflows & monitor all communication channels while maintaining speed and reliability. We used MCP servers to bridge the gap between natural language processing and CRM operations, allowing sales reps to simply type what they want to accomplish rather than navigating through multiple screens and forms.

When creating goatforce I realized our Gmail MCP server was bottlenecking our whole pipeline, I decided to build our own server using fast mcp and created a data dashboard where I could upload our time logs and graph out where our bottleneck could be fixed. After implementing our own MCP server I cut down pipeline time by 30-45 seconds a massive improvement

Since our pipeline would take up to 60 seconds depending on how much new conversation data was added, I needed a way for our users to be able to navigate away from the loading screen / refresh without allowing them access to click into the deal. This is when I implememnted websockets using socket.io to emit a message when our pipeline had started and emit when done, keeping the loading state monitored perfectly

Another issue encountered was a delay in data getting to our web server, everytime the dashboard was refreshed the get request would take 1-2 seconds to load in our assets which is hoestly not that bad but not the ideal user experience. To reduce this I implemented smart redis caching that would cache all of our deal information and monitor if the user had ran a pipeline on a specific deal which would then run a background task to refresh the cache giving us a immensely fast load in time