Hello,
As a learning guide to continue discovering N8n and integration, I am actually trying to integrate Atlassian Rovo MCP server in my workflow in order that from a simple test chat message my AI agent will be smart enought to be instruction and use Rovo as MCP client in order to deliver something relevant to user request.
I have seen from Rovo MCP documentation that it provide different named tool that can be use like createJiraIssue, editJiraIssue,…
My concern is my test is as follow :
Based on user message I need rovo to be able to propose suitable issue type to create (task, incident, bug,…) and then create the jira issue based on user information.
What is unclear for me is how do we instruct the ROVO MCP client to specify the internal tool it should use in order to provide correct information ?
Does is all govern by the AI agent context ?
Thanks for sharing how to clarify this test integration
Hi @scal just configure the MCP client tool with your ROVO MCP end point , to your AI agent node, just make sure to write a very clear system prompt so that AI actually knows what to do, just follow the basics mentioned here on how to set that up:
Thanks for your reply @Anshul_Namdev , yes the basic is actually doen from the begining
The bigest step is the system prompt AI instruction to cover the steps.
My question was more oriented also to the point that is it better to :
1 - create s single MCP client node which is capable to handle tools from MCP server one MCP client 2 - Create a unique MCP client for each MCP server tool we want to use in MCP server
I understand that prompting is the key for sure, and have the feeling sometimes that a small change in prompting for a special thing can break the all thing, so in terms of design what is best
@scal That is a good question, what i would recommend is that it depends on the pattern AI agent would be able to access that , like if it is very linear , like AI would get data, AI would update data.. and something like that which is easier to follow and no or less middle computation is required between operations of tool calling, then you can use single AI agent to manage multiple MCPs , although in production use case consider using multiple AI agent and every agent doing one task at a time would be a better take for less AI made errors in large scale. Also what i use is this tool:
