Skip to main content

MCP

Capture has an MCP server. It gives the AI tools you already use access to the same Capture tools that Captain uses, so you can read and work with your Capture data from Claude, Mistral Vibe, Codex or Microsoft Copilot Studio.


What is MCP?​

The Model Context Protocol (MCP) is an open standard for connecting AI assistants to the tools and data of other systems. It works in three steps:

  1. Discover. The assistant asks a server which tools it offers.
  2. Call. The assistant calls a tool with structured parameters.
  3. Receive. The tool answers with structured data the assistant can reason about.

Before MCP, every combination of assistant and system needed its own integration. With MCP, a system describes its tools once and every assistant that speaks MCP can use them.

Why it matters​

  • Real data, not guesses. An assistant that calls a tool works with the actual value from your system.
  • A tool is a contract. Each tool says what it does, what it needs and what it returns, so an assistant knows when to use it.
  • It is open. You are not locked in to one assistant. Choose the one that suits your team, and switch when a better one comes along.

How Capture uses MCP​

Each function in Capture can define its own tools. They are all registered at one central MCP server, the gateway to everything Capture can do for an AI assistant. It runs behind the scenes, but it is what makes the whole setup work:

  • One door. Captain and your own AI tools use the same server, so a tool is built once and available everywhere.
  • New apps appear by themselves. When a Capture app adds a tool, it shows up for Captain and for your own tools without an update on your side.
  • Permissions are checked for you. You sign in as yourself. A tool is only offered when your role allows it, and a call runs with your permissions.

The tools available today are the ones listed under What Captain can do: the UNS, the raw data explorer, Grafana dashboards, the logbook app, FlowBuilder and AppStudio.


Connect your own AI tool​

What you need
  • Capture Cloud 3.9 or newer.
  • A Public ClientId. This identifies your AI tool to Capture when you sign in. Contact Capture sales to get it.
  • A Capture account whose role has access to the MCP server. Without it, every request is refused with a 403. Ask your administrator to give your role access.
  • An AI tool that supports remote MCP servers.

The address of the Capture MCP server is:

https://mcp-server.apps.captureplatform.com/mcp

You sign in with your Capture account, the same way you sign in to Capture. Your tool asks the server where to sign in, and the Public ClientId tells Capture which tool is asking.

  1. In Claude, open Customize and go to Connectors.
  2. Select + Add, then Add custom connector.
  3. Enter a name, for example Capture, and the address of the MCP server: https://mcp-server.apps.captureplatform.com/mcp. Select Continue.
  4. Under Authentication, choose Sign in now. Under OAuth client, choose Use your own OAuth client and enter your Public ClientId. Leave the secret empty: it is a public client.
  5. Select Add, then Connect, and sign in with your Capture account.

Claude's add custom connector dialog with the name Capture and the Capture MCP server address filled in

Start a new chat in Claude, select + and Connectors, and switch on Capture. Claude can now use your Capture tools.

The labels differ per plan and version

Anthropic changes the screens of Claude from time to time, and what you see depends on your plan. Look for the setting where you add a custom connector and where you can enter your own OAuth client. On a Team or Enterprise plan, an owner of your organization adds the connector first.

Only one group of tools​

The address above gives your tool every tool you may use, in one list. To connect only one group, for example only the Grafana tools, add the name of the group to the address:

https://mcp-server.apps.captureplatform.com/mcp/grafana

Your AI tool can list the groups available to you at https://mcp-server.apps.captureplatform.com/mcp/groups. This request needs your Capture sign-in.


Captain or your own AI tool?​

Both use the same Capture tools. The difference is everything around the tools.

Capture at the left, Captain between Capture and the MCP server, and Claude Code, Mistral Vibe, Codex and Copilot Studio at the right. Every Capture app registers its tools at the MCP server.Capture at the left, Captain between Capture and the MCP server, and Claude Code, Mistral Vibe, Codex and Copilot Studio at the right. Every Capture app registers its tools at the MCP server.
CaptainYour own AI tool, such as Claude, Mistral Vibe, Codex or Copilot Studio
Knows CaptureYes. It knows your companies and what you are looking at.No. It sees a list of tools and has to work out the rest.
GuidanceA native agent framework: specialists per Capture app that know how to use their tools well.A whole bunch of tools, without the guidance on how to combine them.
SecurityBuilt-in. Personal data is redacted before it reaches the model, and every call follows your permissions.Your permissions still apply to every call, but what happens with your data is up to the vendor you chose.
Data processingStrict EU-only processing. Your data is not used for training and is not retained.Under the terms of your AI vendor.
CostLower. Capture selects models that perform well and are good value for money.You pay your AI vendor, with the model you pick.
Where you use itInside Capture, next to your data.In the tool you already work in, for example your editor or a workflow.

Use Captain for everyday questions about your plant, your documentation and your data, and when you want answers you can trust to follow your company and your permissions.

Use your own AI tool when your team already works in one, when you build automations that include Capture data, or when you want to combine Capture with other systems that also speak MCP.


Build on it​

Do you build an app on top of Capture? Your app can add its own tools, and even its own agents and skills, to the MCP server. They then show up in Captain and in your own AI tools. See Build on Captain.