MCP output optimization

Feedback
1

HI!

Rentman’s outputs are very, very, very large. I requested the names of the technicians assigned to a specific function on a specific contract,

To find “sound lead” or “video lead,” the agent first attempted to read the assignments linked to a specific function.

Some list_projectcrew calls on a function only returned planning lines such as:

text

id: 27029

function: /projectfunctions/70694

displayname: Display Planningpersoneel-27029

planperiod_start: 2026-06-06T08:00:00-04:00

planperiod_end: 2026-06-06T18:00:00-04:00

But without the person’s actual name. The agent then made a broader call:

text

mcp_rentman_projects action=list_projectcrew id=6277 limit=300

And there, Rentman MCP returned all crew assignments for project/contract 28996, which is:

text

itemCount: 174

This large return did not contain all readable names, but mostly:

  • assignment IDs;

  • function IDs;

  • crew/crewmember IDs;

  • dates;

  • metadata.

Then, the agent locally filtered the lines that corresponded to the desired function, and then made individual mcp_rentman_crew get calls to convert the crew IDs into names.

The mcp_rentman_crew get calls also returned fields that were useless for this question, notably email and custom. Even if the agent had only requested certain fields, the MCP return still contained more data. This adds tokens and is not ideal for minimization.

Here are some areas for improvement:

  • avoid the broad projects list_projectcrew limit=300 call when a specific function is already known;

  • request as few fields as possible;

  • if the MCP still returns too many fields, add a filter on the wrapper side to keep only the requested parameters;

  • and ideally, never return custom / internal notes / emails to the model unless requested.

Thanks !

2

Thank you for your input, this is a very good point! We are currently working on a fix, I will inform you when you can test it.

3

The fix is now live. Can you please try it again and see the output? It should be working way better, as we added a new parameter to the API endpoints that reduces the need for additional requests.