Does each MCP server need to be dedicated to one tech only (alias Tool), like one MCP server for google drive, one for PostgreSQL, one for docker, etc.? we can have one MCP server handling multiple Tools as well right? I meant, either of the patterns are fine to implement, correct?
Correct, but I imagine some proxy similar to API Gateway that would route through multiple MCP Servers rather than trying to implement all integrations with a single one.
Anyway, your statement is correct. I believe there will be many best practices coming out on how this should be implemented properly.
Got it. I am trying to develop an agnostic MCP server which can be used in any pattern required and also in any language. So, I’m trying to get opinions from experts like you. Thanks for responding.
Agreed. We’re currently following the rule of thumb of the fewer tools per MCP server as possible, based on the same logic that our agent tools have higher correct calling rates when scope/number of tools is as few as possible.
Our product facilitates interoperability between field service businesses that work together, we believe that MCP is the route. The current approach is the proxy you alluded to. The MCP routes the request to the correct agentic flow, leaving the agent/s in that flow to utilise tool that is more granular than the high level abstraction the MCP server provides.
Things like;
Add new job for contractor
Add new customer
Get completed jobs
Add note
Etc
Each of those may have multiple different system access in the agent flow/s but the MCP simply passes the key Params in natural language. Langflow MCP server exposes any flow in your instance as a tool, it’s brilliant and looking very promising!
I hate to be "that guy" but as a write you need to fix up the use of "its" and "it's."
For example, when you write, "[i]n it’s simplest high level definition, an AI agent is an application that uses LLM at the core as it’s reasoning engine to decide on the steps it needs to take to solve for users intent," you use "it's" when you mean "its" -- twice. This is really jarring to some (many? most?) readers.
Does each MCP server need to be dedicated to one tech only (alias Tool), like one MCP server for google drive, one for PostgreSQL, one for docker, etc.? we can have one MCP server handling multiple Tools as well right? I meant, either of the patterns are fine to implement, correct?
Correct, but I imagine some proxy similar to API Gateway that would route through multiple MCP Servers rather than trying to implement all integrations with a single one.
Anyway, your statement is correct. I believe there will be many best practices coming out on how this should be implemented properly.
Got it. I am trying to develop an agnostic MCP server which can be used in any pattern required and also in any language. So, I’m trying to get opinions from experts like you. Thanks for responding.
Agreed. We’re currently following the rule of thumb of the fewer tools per MCP server as possible, based on the same logic that our agent tools have higher correct calling rates when scope/number of tools is as few as possible.
Our product facilitates interoperability between field service businesses that work together, we believe that MCP is the route. The current approach is the proxy you alluded to. The MCP routes the request to the correct agentic flow, leaving the agent/s in that flow to utilise tool that is more granular than the high level abstraction the MCP server provides.
Things like;
Add new job for contractor
Add new customer
Get completed jobs
Add note
Etc
Each of those may have multiple different system access in the agent flow/s but the MCP simply passes the key Params in natural language. Langflow MCP server exposes any flow in your instance as a tool, it’s brilliant and looking very promising!
One of the best resources on MCP out there 🤘
Do you think the technology is mature enough to be adopted in production projects, or should we still sit on it?
I hate to be "that guy" but as a write you need to fix up the use of "its" and "it's."
For example, when you write, "[i]n it’s simplest high level definition, an AI agent is an application that uses LLM at the core as it’s reasoning engine to decide on the steps it needs to take to solve for users intent," you use "it's" when you mean "its" -- twice. This is really jarring to some (many? most?) readers.