I have built an Orchestrator Conversational Agent that provides process-level details such as job status, job state, failure reason, logs, and trigger time.
The issue I am facing is related to process name mapping.
For example, when a user asks:
“What is the status of the Standard Billing process?”
In Orchestrator, the actual folder or process name is SBI.
Currently, I am using a Config file as a context Grounding (for example, “Standard Billing → SBI”) to resolve this. However, this again introduces a maintenance overhead, because every time a new process goes live, I need to update the mapping or context manually.
I am looking for a way to make this more fuzzy or intelligent, so that the agent can automatically understand user intent and fetch the correct process or folder details from Orchestrator without relying on static mappings.
I recently came across the concept of an MCP (Model Context Protocol) server, but I am not very familiar with it.
Could MCP help in this type of use case?
Or is there a better architecture or approach within UiPath to handle fuzzy process identification dynamically?
MCP itself won’t solve fuzzy name matching — it only exposes tools like Orchestrator APIs safely to an agent. The right UiPath approach is to query Orchestrator dynamically for all processes and folders, use AI or fuzzy matching on display names, package names, descriptions, or tags, and cache the results instead of maintaining static config files. MCP can act as the secure bridge to Orchestrator, but the matching logic still sits in the agent.
MCP may not help much here, since the problem is not tool access but semantic matching.
A better approach is to expose a tool that fetches the live list of Orchestrator processes (name, description, tags) and let the agent semantically match the user query (e.g., “Standard Billing”) to the closest process (e.g., SBI).
This removes static mappings and maintenance overhead. MCP makes more sense only when you need cross-system orchestration, not for fuzzy process identification.
MCP alone won’t handle fuzzy process names. A better approach is to fetch all process names from Orchestrator via API and use NLP or string similarity to match the user’s input to the correct process. This removes the need for manual mapping and works dynamically as new processes go live.