Support QA as a tool your OpenAI agents can call.
If you are building a support agent on OpenAI, you need a second opinion on what it says to customers. Supportman is that second opinion, exposed as a remote MCP server: hand it a draft or a finished conversation, get back a structured review.
The review comes back as JSON your code can act on: a 0–100 score across four categories, a predicted CSAT, comments anchored to specific messages, and a suggested rewrite. Gate a send on it, log it, or route low scores to a human.
Supportman for OpenAI is in early preview. The demo is open now, no account needed.
Connect the demo in under a minute.
The demo server is open to anyone. Its reviews are rule-based: instant, free, nothing stored, and no AI on our side. It works in any assistant or agent that lets you add a custom MCP server.
https://supportman.io/mcp
- Step 1
Add https://supportman.io/mcp as a custom connector (MCP server) in OpenAI. If it asks about sign-in, choose no authentication.
- Step 2
Ask: “Show me a sample Supportman review.”
- Step 3
Paste a real thread, an email chain or a reply you are about to send, and ask Supportman to review it.
From Claude Code:
claude mcp add --transport http supportman-demo https://supportman.io/mcp
A code review, for support replies.
Paste or point at any conversation: a Gmail thread, a Slack thread, a helpdesk transcript. Supportman scores it on the same four categories it already uses for AI conversation scoring in Intercom, then tells you exactly what to change. Example below.
The thread you paste
Customer
My order arrived broken. This is the second time. I want a refund.
Agent
Hi, please send a photo of the item and your order number and we will look into it.
Supportman review
69/100 · predicted CSAT 2/5
- Brand voice & tone16/20
- Knowledge & accuracy24/30
- Resolution & helpfulness14/30
- Communication efficiency15/20
- No acknowledgement that this is the second broken order. Lead with that.
- Asks for the order number, which is already on the conversation.
- No outcome offered. Say what happens once the photo arrives.
Suggested reply · I'm sorry, a second broken order is not okay. Send a photo when you can and I'll refund you the same day, no need to return it.
- “Review this draft from our support agent before it sends.”
- “Score the last 20 escalated conversations and list the weakest category.”
- “Return a rewrite for any reply that scores under 70.”
Supportman in OpenAI, one sign-in.
Your own agents
OpenAI models can call remote MCP servers as tools. Point yours at Supportman to review a draft before it reaches the customer.
Codex and ChatGPT
The same server powers the Supportman plugin for ChatGPT, so a team can use it in chat and in code from one connection.
A QA gate for AI replies
Score each draft your bot writes. Send the good ones, hold the risky ones for a person.
Regression checks for prompts
Run a fixed set of tricky conversations through your agent after every prompt change and compare Supportman scores.
The same rubric as your humans
Score bot and human replies against one rubric, so "is the AI as good as the team" has a real answer.
Supportman for OpenAI, answered.
Is this an OpenAI product?
No. Supportman is independent. It connects to OpenAI products through the open Model Context Protocol (MCP).
How do my agents authenticate?
With standard OAuth 2.1 against your Supportman workspace, the same flow every MCP client uses. Tokens are scoped to one workspace and can be revoked from the dashboard.
Is this available today?
A free demo is: add https://supportman.io/mcp to your assistant as a custom MCP server, no account needed. Its reviews are rule-based. The full product, with AI scoring and your Intercom data, is in early preview: request access and we will add you as we open it up, platform by platform.
What does a review score?
The same four categories Supportman already uses for AI conversation scoring in Intercom: brand voice and tone (20), knowledge and accuracy (30), resolution and helpfulness (30), and communication efficiency (20). You get a 0–100 quality score, a predicted CSAT from 1 to 5, line-level comments, and a suggested better reply.
Do I need Intercom to use it?
Not for reviewing a pasted conversation. Connecting an Intercom workspace adds the second half: asking about your real CSAT, DSAT conversations, Fin, and agent performance.
Supportman for OpenAI is in early preview.
Request access and we will let you in as we open the preview up. Already a Supportman customer? Use your work email and we will prioritise your workspace.
OpenAI is a trademark of OpenAI. Supportman is an independent product, not made by, endorsed by or affiliated with OpenAI.
Request early access
Get on the list for Supportman for OpenAI.