Setup (Where it is done)
- Settings → Developers, in app.intake.es
- Generate the MCP server credential
- Choose what the connected client can read and do
Connect your models to Intake conversations over MCP
Model Context Protocol is the standard a model uses to talk to external tools. Intake exposes its own MCP server: any compatible client can read conversations, query the knowledge base and trigger actions with the credentials and permissions you define.
What it solves
Your conversations, available to your own agents
The knowledge support accumulates —what is asked, how it was solved, what breaks most— rarely leaves the support tool. With the MCP server, your models and agents can work with it like any other source, exporting nothing and building no replica.
What it solves
The standard protocol, not another API to learn
Every custom integration is a contract someone has to maintain when it changes. MCP is an open protocol your tools already speak, so connecting Intake is closer to plugging something in than integrating it. And what you learn carries to the next system you connect.
What it solves
You decide how far it reaches
Exposing support conversations to a model is exactly the kind of decision you need to be able to scope. The server exposes what you enable and nothing else, and it is cut from your side without depending on anyone. Starting with public knowledge and widening later is an option, not a workaround.
What it does
What the connection triggers and what the agent does with it
An integration is not useful for being on the list: it is useful for what it lets the agent do. This is what MCP puts on the table, and what the agent does with it without anyone watching.
Actions (What it exposes)
- Read conversations and their context
- Query the knowledge base
- Trigger actions and escalate to a person
How MCP helps the agent resolve with context
Three concrete things that change the team day, not a feature list.
Your agent, our context
If you already have your own agent, there is no need to migrate it: it connects over MCP and works with the conversations and knowledge already living in Intake.
Explicit permissions
The credential defines what the connected client can read and run. Scope is decided when you create it, not at run time.
A standard, not a private format
Because it is MCP, any compatible client works. You do not write an adapter for every tool you want to connect.
How it works
Three steps, and only the first needs a MCP admin. The rest you change later without reconnecting anything.
Authorise the connection
You open the integrations panel, pick MCP and authorise access with the account your team already uses. You need to be an admin in MCP, so if you are not, this is the only step you will have to ask someone for.
You choose what the agent sees
You tick the channels, events and account fields it can reach. What you do not tick does not exist for it, and you can change that whenever you want without reconnecting anything.
It answers with that context
From the first conversation the agent looks that data up before replying, resolves what it can close on its own and passes to a person what it cannot, with the case already assembled.
Frequently asked questions
How long does connecting MCP take?
Minutes, not an afternoon. You authorise access and choose what it can see; nothing needs rewriting and no development work is needed to start.
Can I revoke access later?
Yes, and from either side: revoking the connection in Intake or withdrawing the permission from MCP. The agent stops reading that data immediately and keeps answering with the rest.
Does using MCP cost extra?
No. Every integration sits inside the per-resolved-conversation price: 0.41 EUR on annual billing. There is no add-on to buy and no call meter.
What happens if MCP stops responding?
The agent answers with what it does have and, if the question depended on that data, hands off to a person explaining why. It never invents the value it could not look up.
What is MCP?
Model Context Protocol is an open standard that lets a language model use external tools. Instead of writing one integration per model, you expose an MCP server and any compatible client can use it.
Does Intake expose a server or consume third-party ones?
It exposes its own: your agent or MCP client connects to Intake, not the other way around. The credential and its scope are generated in Settings → Developers.
Can I limit what the connected client sees?
Yes. When you create the credential you choose what it can read and which actions it can trigger. Anything you do not enable stays out of reach.
