"First Form" Implemented MCP to Configure BPM Processes via AI Agent

"First Form" implemented MCP support and demonstrated how a BPM system can be transformed into a functional interface for AI agents. Now, some of the configuration of categories, forms, and processes can be launched not through specification documents and developer queues, but through ordinary natural language dialogue.

The BPM Bottleneck

The problem faced by almost all BPM platforms has long been known: business understands what process they need, but cannot quickly implement it in the system themselves. Even a simple task like a category for vacation approval usually goes through a chain of process owners, administrators, developers, or integrators. At each step, there's a translation from "I know what I need" to "here's how to configure it in the interface and access rights," and along with it—delays and loss of meaning.

Because of this, process configuration turns from a management task into a mini-project with a request, requirements description, and waiting in a queue. If business needs to quickly change form fields, user rights, or routing logic, it depends on those who understand the internal structure of the system. As a result, the BPM that should accelerate company operations becomes a bottleneck itself.

It was precisely this gap between intention and implementation that the "First Form" team decided to bridge through MCP.

How MCP Was Connected

MCP, or Model Context Protocol, exists so that an AI agent can not only respond with text but work with a specific system through described tools and context. In the case of "First Form," this means the agent gains a controlled way to create and configure objects within the BPM: categories, forms, processes, and associated parameters. A user describes the task in ordinary language, and the system translates this into actions that previously were performed manually through the interface or internal settings.

Practically, this changes the very way of interacting with the platform. Instead of a formal specification, you can formulate a request the way business understands it: what entity is needed, which fields should be mandatory, who has access, and how approval flows. If the integration is built cleanly, the agent becomes not a "chat next to the system," but a full-fledged management layer.

According to the team, they haven't seen ready-made MCP server implementations in Russian BPM systems yet, so the case looks not like an experiment for hype, but as an infrastructural shift.

What This Gives

The main effect here isn't that AI writes configurations instead of humans, but in reducing the distance between process idea and its launch. The fewer intermediaries between the process owner and the working configuration, the faster the company tests hypotheses and makes changes. For a BPM platform, this is especially important because business processes rarely remain stable: they are regularly supplemented, simplified, or rebuilt under new rules.

• creating categories by text description
• configuring fields and forms without manual assembly
• assigning rights and roles through natural language
• launching changes without a long specification cycle and queues

Separately important is that we're talking not about generating pretty mockups or drafts, but about access to real actions within a corporate system. This makes the MCP topic much more practical: value appears where the agent can safely change the working environment according to clear rules. If such a model takes hold, BPM vendors will have a new standard of expectations: users will want not just an interface for administrators, but a conversational management layer on top of it.

What This Means

The story of "First Form" shows that AI agents are gradually transforming from text assistants into an interface for corporate software. If MCP truly becomes a standard of integration, business will be able to configure internal processes considerably faster, and the role of developers will shift from manual configuration to control, rules, and architecture.