Interfaces · Claravine MCP
Your taxonomy, inside the AI tools your teams already use
Connect Claude, ChatGPT or any MCP-capable assistant to Claravine, and your team can find templates, validate data and submit campaigns by describing what they want — against the governed standard, not a copy of it.
Trusted by
Why the default way runs out
An assistant is only as good as what it can see. Give it a screenshot of your taxonomy and it guesses; give it the taxonomy itself and it complies.
- You pasted your taxonomy into a chat window, and the moment it changed the assistant was working from a stale copy.
- Your team asks the AI for a campaign name and gets something plausible that fails validation the moment it is submitted.
- The people who most need the standard are the ones least likely to open the tool that holds it.
What your team can do with it
Find the right template by describing it
“Find my Social Media Campaign template and show me what fields I need to fill out.” No navigating, no remembering field names.
Validate before anything is submitted
“Validate this data against my Ads template.” Missing required fields, invalid values and formatting problems surface before they reach a platform.
Submit in plain language
“Submit this file to the Campaigns template. Validate the values first.” The assistant checks against your governance rules, then submits.
Ask what the rules actually are
“What values are allowed in the Region field of our Campaign template?” New joiners ask the assistant instead of reading documentation or booking training.
How it works
Claravine runs an MCP server. MCP is an open standard — think of it as a common plug that lets an AI assistant talk to a system directly rather than through copy-paste. Point a compatible client at the server, authenticate, and the assistant gets a defined set of tools: reading your templates, checking values against your lists, validating rows and creating submissions. It works through the same rules your web interface enforces, because it is the same rules.
Nine tools are exposed, in three groups.
Connecting it, in three steps
No configuration file to write. If your client supports MCP, this is the whole setup.
- 1
Pick a client.
Claude and ChatGPT connect over OAuth; Goose accepts either OAuth or an API key. Any MCP-capable client works — the protocol is open, so a new one works without anything changing here. - 2
Point it at the server.
Transport is Streamable HTTP. OAuth clients usehttps://mcp.claravine.com/oauth/mcp; API-key clients usehttps://mcp.claravine.com/mcpwith anx-claravine-keyandx-claravine-secretheader pair. - 3
Sign in and ask.
OAuth walks the user through signing in with their Claravine credentials. From there the assistant can list templates, check allowed values, validate rows and submit — as that user, within that user’s permissions.
What it can see, and what it cannot
- The assistant acts as a user, not above one. It sees what that user’s role and permissions allow, and nothing else.
- Two authentication methods — an API key, or OAuth. OAuth does not require the public API-key feature to be enabled.
- Every submission made this way is subject to the same validation and the same activity log as one made in the interface.
- Running on SOC 2 Type II infrastructure with multi-tenant isolation, SSO and MFA.
The same taxonomy, whichever way you reach it
These are not three products. They are three ways into the same governed taxonomy — the one your templates define, your approvals control and your teams already submit against. A change approved in the interface is the change the API returns and the agent works from. Nothing forks, and nothing has to be kept in sync by hand.
82
Source: Claravine call corpus, classified
Documentation
Setup, authentication, client configuration and the full tool reference are in the product documentation. There is no configuration to read on this page — if you are the person who will wire it up, start there.
Frequently asked questions
Didn't find your answer? Bring your questions to a thirty-minute walkthrough with someone who has done this at enterprise scale — we'll show you, not tell you.
What is MCP?
The Model Context Protocol is an open standard for connecting AI assistants to external systems. Before it, every integration was custom-built. With it, an assistant can reach a system through one common interface — which is why the same Claravine connection works across different AI tools rather than needing one build per tool.
Which AI tools does this work with?
Any client that supports MCP. Claude and ChatGPT are the common ones today, and the list grows as the standard is adopted. Because the protocol is open, a new client that supports MCP works without anything changing on our side.
Can the assistant see data it should not?
No. It acts as an authenticated user and is bound by that user's role and permissions. It cannot read a template someone is not entitled to, and it cannot submit past a validation rule. If a person could not do it in the interface, the assistant cannot do it on their behalf.
Do we need the API for this?
Not necessarily. API-key authentication requires the public API-key feature; OAuth does not. Which you choose depends on how your organization prefers to handle credentials, not on what the assistant can do.
Is this a replacement for the web interface?
No, and it is not meant to be. It suits the people who work in an assistant all day and the tasks that are faster to describe than to click. The interface remains the place to design templates and manage governance.
See it against your own taxonomy
Thirty minutes with someone who has implemented this at enterprise scale.