How to Use the VideoSupport MCP Server: Give Your AI Support Agent Video Context
Learn how the video support MCP server gives your AI agent rich visual context to triage issues, request recordings, and resolve tickets faster.
If you already run an AI support agent, you've probably hit a familiar ceiling. The agent can query your order system, check warranty status, and draft replies — but it can't see what the customer is actually looking at. Someone types "my router keeps blinking orange" and the agent has to guess: firmware fault, loose cable, or a hardware failure that needs an RMA?
The VideoSupport MCP server closes that gap. It gives your agent a set of tools to request a customer video, read the transcript, pull a visual triage, and reply or escalate — all without leaving your existing workflow. This guide covers what the MCP server does, how to connect it, and how to get the most out of it.
What MCP Is and Why It Matters for Support
MCP (Model Context Protocol) is an open standard created by Anthropic in November 2024. It defines a common way for AI applications to connect to external tools and data sources, so any MCP-compatible client can talk to any MCP-compatible server using the same protocol.
The VideoSupport MCP server is a remote server — meaning you connect via a URL rather than installing anything locally. Your agent authenticates with an API key, calls the tools it needs, and gets structured data back. No custom integration code, no webhook plumbing.
The Six Tools Your Agent Gets
Once connected, your agent has access to six tools:
| Tool | What it does |
|---|---|
request_recording(ticket_id, prompt) | Sends the customer a recording link tied to a specific ticket, with a prompt guiding them on what to show |
get_recording(video_id) | Retrieves the recorded video once the customer has submitted it |
get_transcript(video_id) | Returns the full text transcript of the recording |
get_visual_triage(video_id) | Returns VideoSupport's visual analysis: detected devices (routers, printers, monitors, laptops, phones), observed symptoms, and a likely issue |
reply_to_customer(ticket_id, message) | Posts a reply to the customer inside your connected helpdesk |
escalate_to_human(ticket_id, reason) | Hands the ticket to a human agent with the video, transcript, and triage notes already attached |
The customer records in a browser on any device — phone, tablet, or desktop. No app download, no browser extension, no friction.
The End-to-End Flow
Understanding the sequence helps you design your agent logic correctly.
Step 1: Your agent requests a recording. When a ticket arrives that needs visual context, your agent calls request_recording. The prompt parameter is where you tell the customer exactly what to show — for example: "Please show us the front of the router, the lights that are blinking, and any cables connected to the back."
Step 2: The customer records. They click the link, record in their browser, and submit. No install required on their end.
Step 3: VideoSupport triages the video. VideoSupport processes the recording and produces a transcript and a visual triage that identifies devices in frame and surfaces a likely issue. Your agent can then call get_transcript and get_visual_triage to read both.
Step 4: Your agent applies your business logic. This is where your own tools come in. The agent might check an order number, verify warranty status, look up a known fault code, or determine whether the symptom matches a replaceable part. VideoSupport surfaces a likely issue — the final diagnosis stays with your support team or your agent's logic. Don't treat the visual triage as a definitive verdict.
Step 5: Reply or escalate. If the agent can resolve the ticket, it calls reply_to_customer. If the issue needs a human, it calls escalate_to_human with a reason, and the human agent receives the ticket with the video, transcript, and triage notes already attached. Replies land in HubSpot, Intercom, Zendesk, Crisp, or Monday.com.
A Worked Example: Router Blinking Orange
Here's how that flow looks in practice.
A customer opens a ticket: "My router has been blinking orange for two days and I have no internet."
Your agent recognizes this as a hardware-or-config ambiguity and calls request_recording with a prompt like: "Show us the front panel of your router, the blinking lights, and the cable going into the WAN port."
The customer records a 30-second clip. VideoSupport's visual triage comes back: device identified as a home router, observed symptom is a solid orange WAN light, likely issue is no upstream signal or a failed WAN port.
Your agent calls get_visual_triage, reads the output, then queries your warranty system using the model number visible in the transcript. The router is 14 months old and inside the two-year warranty window.
Now the agent has enough to act. If your logic says a failed WAN port on a warranted device gets an automatic RMA, the agent calls reply_to_customer with the RMA instructions. If your policy requires a human to approve hardware replacements above a certain value, the agent calls escalate_to_human with the reason "warranted hardware fault, RMA approval needed" — and the human sees the full video context immediately.
Either way, the first reply contains the right answer. No back-and-forth asking the customer to describe the lights again.
How to Connect the MCP Server
MCP is available on Pro and Enterprise plans. To connect:
- Go to mcp.videosupport.io for current connection details and client configuration instructions.
- Generate a per-workspace API key inside your VideoSupport account.
- Connect a remote MCP-capable client — compatible options include Claude, Cursor, VS Code, or your own agent framework.
- Follow the exact configuration steps at mcp.videosupport.io for your specific client. That page is kept up to date, so it's the right place to pull config details rather than relying on anything reproduced here.
Every action your agent takes through the MCP tools is logged on the ticket, giving you a full audit trail of what the agent requested, what it received, and what it sent.
One billing note worth knowing: recordings that your agent handles end-to-end via MCP don't count against your monthly AI review quota. You're using your own agent's intelligence; VideoSupport is providing the video capture and triage layer.
When to Use MCP vs the Built-In AI Review
VideoSupport has a built-in AI review that reads incoming recordings and drafts a reply without any agent integration. That's the right choice when you want coverage out of the box and don't have an existing agent to connect.
The MCP server makes more sense when:
- You already have an AI agent with access to your order system, warranty database, or product catalog
- You want the agent to decide whether to reply or escalate based on your own business rules, not a generic model
- The agent's reply needs to draw on data VideoSupport doesn't hold — a customer's purchase history, an open RMA, a custom pricing tier
- You want every agent action logged in your own audit trail
If you're starting from scratch and want AI-handled video tickets without building an agent, the built-in review is the faster path. If you have an agent and want to give it eyes, MCP is the right tool.
Best Practices
Write specific request_recording prompts. A vague prompt produces a vague recording. Tell the customer exactly what to show, in what order, and from what distance. "Show the front of the device, then the back panel, then any error message on screen" produces far more useful footage than "show us the problem."
Set clear escalation rules before you go live. Decide in advance which ticket types your agent is authorized to resolve and which always go to a human. Hardware faults above a cost threshold, safety-related issues, and complaints from high-value accounts are common candidates for automatic escalation. Build those rules into your agent logic from the start, not as an afterthought.
Start with low-risk ticket types. Before routing your entire queue through the agent, run it on a category where a wrong reply is easy to correct — how-to questions or shipping status, for example. Review the logs, check the escalation rate, and tune your prompts before expanding scope.
Review the audit logs regularly. Because every agent action is logged on the ticket, you can see exactly where the agent got it right and where it missed. That data is how you improve your prompts, your escalation thresholds, and your tool-calling logic over time.
Don't skip human-in-the-loop for ambiguous cases. VideoSupport's visual triage is a strong signal, not a guarantee. When the triage returns a low-confidence likely issue or the transcript is unclear, your agent should escalate rather than guess. A customer gets a faster, more accurate answer from a human with full video context than from an agent working with incomplete information.
According to VideoSupport's own data, around 75 percent of tickets are resolved without human review when the full video-plus-agent workflow is in place. That figure is not independently verified, but it reflects the direction of travel: fewer escalations, faster first replies, and less time spent asking customers to describe what they can simply show.
Teams like Cowboy (electric bikes) and Print.com are quoted on the VideoSupport site as using the platform — which gives a sense of the hardware and product-support contexts where visual context makes the biggest difference.
Frequently Asked Questions
Does the customer need to install anything to record a video? No. Recording happens entirely in the browser. The customer clicks the link your agent sends and records on any device — phone, tablet, or desktop — without downloading an app or extension.
Which helpdesks does VideoSupport integrate with?
Replies sent via reply_to_customer land in HubSpot, Intercom, Zendesk, Crisp, and Monday.com.
Do MCP-handled recordings count against my monthly AI review limit? No. When your own agent handles the ticket end-to-end via MCP, those recordings don't use your monthly AI review quota.
Can I use the MCP server with my own custom agent framework? Yes, as long as your framework supports remote MCP connections. Claude, Cursor, and VS Code are confirmed compatible clients. For custom frameworks, follow the connection instructions at mcp.videosupport.io.
What happens when the agent escalates a ticket? The human agent receives the ticket with the video, transcript, and visual triage notes already attached. They don't need to ask the customer to re-explain anything.
Is the visual triage a definitive diagnosis? No. VideoSupport identifies devices and surfaces a likely issue based on what's visible in the recording. Final diagnosis — and any decision to replace hardware, issue a refund, or take another action — stays with your support team or your agent's own business logic.
Where do I find the API key and connection details? Go to mcp.videosupport.io. Your API key is generated per workspace inside your VideoSupport account.
Connect Your Agent Today
If your AI support agent can already query orders and check warranties, adding video context is a natural next step. The VideoSupport MCP server gives your agent the tools to request a recording, understand what it shows, and act on it — without replacing your helpdesk or your team.
Connect at mcp.videosupport.io and start a free trial at videosupport.io.