Can I work with my training results through MCP?

What a connected AI assistant can read from your Coassemble results, and the routes for getting that data into your other systems.

An AI assistant connected to your workspace over MCP (the Model Context Protocol) can read your Coassemble results, including course and screen level analytics, so you can ask it questions about how your training is performing. Results are read-only over MCP. To move that data into another system, you use a CSV export, the tracking API, or webhooks.

Connecting an assistant

The connection is made once, from Settings, and authorised with a single sign-in to Coassemble rather than an API key you paste in. Once connected, the assistant has over 50 tools available to it, covering course creation, editing, publishing and analytics. Reading results is one part of a much wider set.

You can see every assistant you've authorised in Settings, and disconnect any of them at any time.

What a connected assistant can read

An assistant can pull the same results you see in Insights: views, completion, drop-off, sentiment, and per-person activity for people you've identified, at both course and screen level. That means you can ask which screen people are dropping off at, or how a quiz is scoring, and get an answer without opening a report yourself.

It reads what you could read. A connection is scoped to the workspace you authorised it from and can't reach anything you couldn't reach.

Getting results into your other systems

The routes for moving Coassemble data into a CRM, HR system, data warehouse or internal dashboard are:

• CSV export. Download the data from the Insights view for the date range you've selected, then work with it wherever you like.
• Tracking API. Pull views, progress, completion and quiz data programmatically so results flow into your own tools and stay current.
• Webhooks. Receive a server-to-server message the moment a course is created, started or completed, so your systems stay in sync without polling.
• In-page events. Listen for events while a course is embedded in your own product.

If your assistant is also connected to the system you want the data in, it can carry results across for you. What it can reach on that side depends on your assistant, not on Coassemble.

Driving this from your own product

If you deliver Coassemble inside your own product or platform, your application can drive the same agent over MCP or REST rather than a chat assistant doing it, using the client and user identifiers your integration already uses.

Three controls matter here, and one of them is directly relevant to results:

• Read-only mode. Let the agent read and answer questions without being able to change anything. If what you want is analytics questions answered inside your product, this is the setting that gives you exactly that and nothing more.
• Course scoping. Limit it to a single course.
• Tool selection. Choose which tools are available rather than exposing all of them.

What is MCP?

MCP, the Model Context Protocol, is an open standard for connecting AI assistants and agents to external data and tools. It lets an AI tool read your Coassemble results through a documented, permissioned connection rather than a one-off custom integration.

Frequently asked questions

Can a connected assistant change or delete my results?

No. Result data is read-only over MCP. A connected assistant can create and edit courses, but it reads your analytics rather than writing to them.

Which assistants can do this?

Any assistant that supports connecting to a remote MCP server and signing in. MCP is an open standard rather than a private integration, so this isn't a fixed list, and support depends on the assistant rather than on Coassemble.

Do I still need the API if I have an assistant connected?

It depends what you're doing. Ask an assistant when you want an answer or a one-off pull. Use the tracking API or webhooks when you want results flowing into another system automatically and on a schedule.