Keragon ships three products that automate the work of running a healthcare business: Workflows, Agents, and MCP. They sound similar, but each one solves a different problem — and knowing which to reach for saves you a lot of head-scratching later. This article explains the difference in plain English.
The short version
Workflows
An automation you design step by step. You wire up each action, and it runs the exact same way every time. Fully deterministic.
Best for
Predictable, repeatable processes where the steps never change.
Example
When a patient books in Acuity Scheduling, create a contact in HubSpot and send a welcome SMS via Spruce Health.
Agents
An AI teammate you brief like a new staff member. You describe the goal in plain English, hand over the tools, and the agent figures out how to get the job done — even when inputs vary.
Best for
Tasks that need judgment, or where the work changes run-to-run.
Example
When a new patient submits an intake form, read the free-text chief complaint, decide whether the symptoms suggest urgency, and route urgent cases straight to the on-call provider.
MCP
A bridge that lets your favourite AI tool — Claude Desktop, ChatGPT, Cursor — talk to your healthcare apps through Keragon's connector library.
Best for
Ad-hoc, conversation-driven work in the AI tool you already use every day.
Example
In Claude: "pull last week's no-shows from our scheduler and draft outreach for each one."
Rule of thumb. Workflows when the how never changes. Agents when the how depends on what's in the data. MCP when you want your AI tool to act on your apps in the moment.
Workflows
A workflow is an automation you design step by step. You pick a trigger, you connect a sequence of actions, and every time the workflow runs it does exactly the same thing in exactly the same order. Workflows are deterministic — if you set one up to send an SMS after a form is submitted, that SMS goes out every single time, without variation.
Use a workflow when
- The same exact steps run every time.
- You want full control over what happens at each step.
- The work is straightforward routing, syncing, or notifying — no decisions to make.
Agents
An agent is an AI teammate you give a job to. Instead of mapping out every step yourself, you write a plain-English brief — the way you'd brief a new staff member on their first day — telling the agent what you want done, giving it the tools it needs, and setting the rules for how to behave. Each time the trigger fires, the agent decides which steps to take, including how to handle the messy cases that don't follow a clean script.
Use an agent when
- The task needs judgment (e.g. "is this complaint urgent?", "what is this message really about?").
- Inputs vary run-to-run — free-text fields, ambiguous data, edge cases.
- A workflow would need a dozen branches to cover every scenario.
Workflows and Agents are designed for different jobs, not as alternatives. Workflows give you full control when the how never changes. Agents take over when the how depends on what's in the data. Most healthcare teams that run Keragon end up using both — workflows for the predictable rails, agents for the judgment calls.
MCP
MCP stands for Model Context Protocol. Keragon's MCP product lets the AI tools you already use — Claude, ChatGPT, Cursor, Mistral, Intuist.ai — talk to your healthcare apps through Keragon's connector library. Instead of building automation inside Keragon, you bring your own AI and give it the keys to your tool stack.
Use MCP when
- You already work in an AI chat tool every day and want it to act on your real data.
- The work is conversation-driven and ad-hoc — "pull this patient's last 5 appointments from DrChrono."
- The task is exploratory, one-off, or research-style — not a repeatable trigger-based process.
Learn more at keragon.com/mcp.
Which one should I pick?
Pick a Workflow when…
- The same exact steps need to run every time, the same way — no judgment involved.
- You want full control over each step in the sequence.
- The work is straightforward routing, syncing, or notifying.
Every Monday at 7 AM, pull last week's appointment count from DrChrono and log it to your reporting sheet.
Pick an Agent when…
- Reading free text or making a judgment is part of the job (urgency, sentiment, intent, fit).
- Inputs change run-to-run and a workflow would need many branches to cover every case.
- You want to brief the work once and have the agent handle the messy or unpredictable cases.
Read each new patient enquiry, classify it (insurance question, booking, clinical concern), and draft a response in the right tone for each category.
Pick MCP when…
- You spend your day in an AI chat tool (Claude, ChatGPT, Cursor) and want it to act on your real apps.
- The work is ad-hoc and conversation-driven — not a recurring trigger.
- The task is exploratory or one-off — data lookups, drafting, research — rather than a repeatable process.
Ask ChatGPT — "pull the patients who cancelled last week and draft a re-engagement message for each one."