AI Agent: API/MCP to create and edit AI Agents
under review
K
Kantu Dev Smartbot
Business Problem
Businesses that build AI Agent setups for multiple clients on respond.io must configure every agent manually through the UI — creating agents, pasting prompts, setting up variables, and repeating this for each specialist agent in a multi-agent setup. Each client engagement takes days to over a week, and every requirement change means editing agents one by one in the UI.
This doesn't scale. Managing multiple clients in parallel turns the UI into a bottleneck, manual copy-paste increases configuration errors, and there's no way to version, template, or programmatically replicate agent configurations across clients.
Desired Outcome
• Expose API / MCP endpoints to create, read, update, and delete AI Agents and configurations including prompts, instructions, variables, and settings.
• This would let users automate agent provisioning, maintain configurations outside the UI, and integrate with external AI orchestrators to generate and push agent setups into respond.io without manual intervention.
Use Cases
• A services team onboarding a new client - With programmatic access, an external agent or script could take client requirements as input, generate the prompt and configuration, and provision all three agents automatically — reducing setup from days to hours and making iteration near-instant.
Current Workaround
N/A
S
Shi Hui
updated the status to
under review
Digi Arabia Marketing Team
Feature Request: Workflow Management via MCP + AI-Assisted Implementation
🔧 THE REQUEST
We've built a comprehensive automation architecture on Respond.io — covering appointment confirmations, reminders, no-show follow-ups, lab results, satisfaction surveys, retargeting logic via Google Sheets, and cross-department patient routing across 9 medical specialties.
The system is deeply integrated with our HIS (Hospital Information System), and we're continuously iterating — adjusting workflows, adding departments, and refining logic as the system scales.
The current gap: Workflow creation and editing is only possible via the visual UI builder. This means every iteration — even a small logic change — requires manual rebuilding inside the interface.
What we're requesting:
- Workflow Management Tools in the MCP Server
The existing MCP server is excellent for contacts, messaging, and tags — but it doesn't expose Workflow CRUD operations. Adding tools like:
- create_workflow
- update_workflow_step
- list_workflows
- clone_workflow
...would allow AI assistants like Claude to build, modify, and manage Workflows programmatically.
- Native Claude / AI Integration
The ability to connect Claude directly to a Respond.io workspace — so an AI assistant that fully understands the business logic (patient journeys, department routing, message timing) can implement and iterate on Workflows without requiring manual UI work.
---
💡 WHY THIS MATTERS
I currently work with Claude (Anthropic) as my implementation partner — Claude has full context of our patient journey architecture, HIS integration design, retargeting logic, and all custom field mappings. The bottleneck isn't the thinking or the design — it's the manual execution inside the UI builder.
With MCP Workflow support, what currently takes hours of careful UI work could be done in minutes through natural language — and iterated in real time as business needs change.
This isn't just useful for us — any agency, healthcare operator, or enterprise client managing complex multi-step automations would benefit enormously.
---
I'd love to discuss this further or contribute to a beta program if one exists. Happy to share our full architecture as a reference use case.
Thank you for building such a powerful platform — the foundation is already excellent.
Best regards,
Khaled
Respond.io Partner — DMC Medical Complex
Al-Madinah Al-Munawwarah, Saudi Arabia
A
Alyaa See
Hi Digi Arabia Marketing Team thank you for the detailed write up.
Your request around Workflow management via MCP is a distinct ask from the AI Agent API request on this post, so we've logged it as a separate feature request here.
We'll keep track of it and make sure to keep you updated on any progress!
A
Alyaa See
Hi Kantu Dev Smartbot, thanks for the suggestion.
To better understand the use case, could you share a bit more about why you need to manage agents and workflows outside the UI, what process you’re handling manually today, and what problems that creates for your team?
An example of your current workflow and where it breaks can be really helpful!
K
Kantu Dev Smartbot
Hi Alyaa See, Sure, here’s more context on the use case and why we need to manage agents (and ideally workflows) outside the UI.
### Why we need to manage agents outside the UI
I work on projects where each client needs a
fully customized AI Agent
(or multiple specialist agents) on respond.io. For each new client, I have to:- Gather their requirements (use cases, tone of voice, restrictions, escalation rules, etc.).
- Design and write the
prompt
for the main AI Agent.- Sometimes create
additional specialist agents
for specific topics or workflows.- Configure
variables, settings and testing flows
around these agents.Doing all of this
manually in the UI
works, but it doesn’t scale: it becomes a bottleneck when you want to deliver a full AI solution in less than a week or when you manage multiple clients at once.What I’d like instead is to build an
external autonomous agent
(e.g. using Claude, Gemini, OpenAI, etc.) that can:- Take the client’s requirements as input.
- Design and refine the prompt(s).
-
Create and configure AI Agents in respond.io automatically via API/CLI/MCP
, instead of me having to click through the UI every time.This turns the process into something repeatable and automatable, instead of manual configuration per client.
K
Kantu Dev Smartbot
### What process I’m handling manually today
Today, for each new AI Agent project I have to:
1. Read the client’s requirements.
2. Manually draft and refine the prompt in a separate tool (or directly in respond.io), going back and forth.
3. Go into respond.io UI and:
- Create a new AI Agent.
- Paste the prompt.
- Configure basic settings.
4. If needed, repeat the same for
secondary/specialist agents
.5. Create or adjust
variables
and any related workflows or logic around those agents.6. Test, tweak the prompt, go back to UI, update again, and so on.
All of this is done by hand, and each iteration requires manual changes in the UI.
### Problems this creates for the team
-
Time-consuming:
Setting up and iterating on an agent (or multiple agents) can easily take more than a week
per client when you include all refinements.-
Hard to scale:
If we want to onboard several clients in parallel, the manual UI work becomes a serious bottleneck.-
Error-prone:
Copy-pasting prompts, recreating similar settings for multiple agents, and keeping everything in sync increases the chance of configuration errors.-
Hard to version/control:
There’s no easy way to treat agent configuration as “code” (e.g. track changes, reuse templates, clone setups programmatically).K
Kantu Dev Smartbot
### Example of current workflow and where it breaks
Example:
1. A new client wants a
multi-agent AI setup
: one general support agent and two specialist agents (e.g. billing and technical).2. I collect requirements and design the structure of those agents.
3. I manually:
- Create the main AI Agent in respond.io.
- Write and tweak the prompt.
- Create the two specialist agents and their prompts.
- Configure variables and any logic needed around their behavior.
4. After testing, the client requests changes (tone, policies, additional constraints, new skills).
5. I must
go back into the UI and edit each agent manually
, re-test, and repeat the cycle.Where it “breaks” is when this needs to be repeated across many clients or when the client’s requirements change frequently. The manual UI work doesn’t keep up with the speed at which we can iterate using external AI tools.
If there were an
API/CLI/MCP to create and update AI Agents
, I could:- Use an external autonomous agent to generate and maintain prompts and configurations.
- Automatically push those configurations into respond.io.
- Reduce setup from
days/weeks
to hours or minutes
, and make the whole process more reliable and scalable.That’s why managing agents (and ideally related workflows) outside the UI is so important for my use case.