From 10–15 Minutes of Setup to a Chat Message
A small example of where MCP is already making a massive difference in AlgoDesk. Before, if I wanted to run a WFO test, there was a fair bit of boring setup involved: create the WFO configuration, create the periods, plug in the strategy, select the correct data, set the increments for every parameter I wanted to optimize, double-check everything and finally run it.
Nothing particularly complicated, but easily 10–15 minutes just to prepare the test. Now I can do basically the same thing from chat in under 30 seconds.
The Bigger Change Is Persistence
The more interesting part is not actually the speed. It is the fact that the whole research workflow becomes persistent.
Last week I was working on a strategy and ran what I call an Edge Survey on it. Basically, I took the strategy with mostly default parameters and ran it across around 25 different symbols just to see where it naturally behaves well and where it doesn't. Then I moved on to something else.
Today I came back to it. Instead of opening files, checking previous tests, looking through results and trying to remember what the hell I was doing with this strategy, I simply asked Claude what we tested, what the results were and what we decided to do next.
It reminded me that the strategy performed quite well on MES and that we decided the next step would be to take it into WFO.
Continuing Exactly Where I Left Off
From there we just continued the research. We discussed what period would make sense, which parameters should actually be optimized, what ranges to use and how the WFO should be structured.
Then I simply told it to set the test up for me. Done.
At that point it can either run the test itself through the AlgoDesk MCP server, or I can still open AlgoDesk and run it manually if I want to watch what is happening.
That is the part I find much more interesting than simply saying "AI saves time." Yes, it saves time. Setting up something that previously took 10–15 minutes can now take less than 30 seconds. But more importantly, the research process itself stops being a collection of disconnected sessions.
The Research Context Stays With You
The previous tests stay there. The decisions stay there. The results stay there. And when I come back a week later I am not starting from zero trying to remember why I tested something, what I liked about it and what I was planning to do next.
It becomes much closer to having another researcher sitting next to you who actually remembers the work.
For systematic trading this is a pretty big deal. The workflow is no longer just:
Idea → Code → Test
It becomes something much more continuous:
Idea → Code → Edge Survey → Decide Next Step → WFO → Sim → Live → Monitor
And MCP basically connects all of those stages together. That, for me, is where this starts becoming genuinely useful.