The Fullview MCP server lets your AI assistant read your Fullview session data, so you can ask about a session in plain language instead of clicking through the dashboard. Ask what went wrong in a session and the assistant pulls the summary, the user journey, the network events even the audio transcript when it's available.
MCP (Model Context Protocol) is an open standard for connecting AI assistants to the data they need. Fullview runs an MCP server that you can connect to Claude and most other assistants that support it.
Where to find it: go to Settings → Installation and scroll to the Connect Fullview to your AI assistant section. Everything you need to connect is there.
Before you start
MCP access has to be enabled for your organisation. If you can't find the Connect Fullview to your AI assistant section, it probably isn't enabled yet — email us and we'll turn it on.
You need an AI assistant that supports MCP. Claude Code, Claude Desktop and claude.ai all do, as do Cursor, Codex, ChatGPT and others.
How to set it up
The Connect Fullview to your AI assistant section has a tab for each kind of assistant. Pick yours and copy what it gives you — the address is already filled in for your account.
Claude Code
Copy the command from the Claude Code tab and run it in your terminal.
Start a Claude Code session and type
/mcp.Select Fullview, then choose Authenticate.
Log in in the browser window that opens, exactly as you would in the dashboard.
Claude Code shows the server as connected once you're done.
Claude Desktop and claude.ai
Copy the address from the Claude Desktop tab.
In Claude, go to Settings → Connectors → Add custom connector.
Name it Fullview and paste the address.
Click Connect and log in with your Fullview account.
Cursor, Codex and other assistants
Copy the configuration snippet from the third tab.
Paste it into your assistant's MCP settings.
Follow the prompt to log in with your Fullview account.
To check the connection, ask your assistant to use the Fullview ping tool. A reply naming Fullview means everything is working.
What you can ask
You don't call the tools by name. Describe what you want, and the assistant picks the right tools and chains them together. Prompts that work well:
Debug a single session.
Session <session ID>: what happened, and what went wrong? Check the summary first, then walk the events and call out any failed network requests or console errors.
Find sessions, then dig in.
Find sessions from the last three days that had rage clicks, then tell me what the most frustrated ones have in common.
Chase a specific failure.
Session <session ID>: list every failed network request with its status and timing, and tell me what the customer was doing right before the first one.
Match a call to what was on screen.
Session <session ID>: interleave the transcript with the events. When the customer says something went wrong, what was happening on screen at that moment?
Compare sessions.
Sessions from last week longer than five minutes with console errors — is there a common page or browser?
You'll find a session ID at the end of the dashboard URL when you open a replay.
What each tool does
Tool | What it gives your assistant |
| Finds sessions by date range, customer, agent, duration, page visited, or signals like rage clicks and console errors |
| Timing, device and browser, who took part, and the frustration signals counted for the session |
| The AI summary of the session and its outcome |
| A readable timeline of what the user did, plus failed network requests, console errors and periods of inactivity |
| The transcript of a call recording, with speakers and timings that line up with the timeline |
| Checks that the connection works — useful when something looks wrong |
Good to know
It's read-only. No tool can change, delete or create anything in Fullview.
You only see what you'd see in the dashboard. The connection uses your own Fullview login, so your assistant works within your normal permissions, and only within your organisation.
Summaries have to exist already. Your assistant reads the AI summary of a session but never generates one. If a session doesn't have one yet, open it once in the dashboard and ask again.
Transcripts need audio. Only cobrowse recordings captured with audio have one, and transcription finishes a few minutes after the call ends.
Searching for one person by email doesn't work directly. Ask by name instead, or use a free-text search — it covers name, email and external ID.
Anything the tools return goes to your AI assistant. Session contents, customer names and transcripts come back as they are, so treat them like any other customer data your team handles.
Notes
The login renews itself, so you only log in once per assistant. After roughly a week without using it, you'll be asked to log in again.
If your account belongs to more than one Fullview organisation, you'll choose one during login. To switch later, disconnect the server in your assistant and connect again.
Long answers can get cut off by your assistant. If that happens, ask for a shorter timeline or fewer transcript segments.
If you have any feedback or question, don't hesitate to reach out to us on support@fullview.io.