Browse docs
Getting Started
Channels
Tickets
AI
Workflows
Teams
Reports
Integrations

MCP Connections

What is MCP?

MCP (Model Context Protocol) is an open standard that lets AI assistants call tools provided by external services. In There There, MCP connections extend what the AI chat assistant can do beyond searching documentation and managing tickets.

For example, connecting your error tracking service lets the AI look up recent errors when a customer reports a bug. Connecting your billing system lets the assistant check a customer's subscription status while drafting a reply.

How it works

An MCP connection points to an external server that exposes a set of tools. When the AI assistant needs information it cannot find in your brain or ticket history, it can call these tools automatically. You do not need to tell the assistant which tool to use. It reads the tool descriptions and decides when they are relevant to the conversation.

Each MCP connection belongs to a brain, and the tools it provides are available to the AI assistant whenever that brain is linked to a channel.

Adding an MCP connection

  1. Go to Settings > Brains and select the brain you want to extend.
  2. Open the Integrations tab and click to add a new connection.
  3. Provide a name for the connection (e.g., "Flare Error Tracking").
  4. Enter the URL of the MCP server.
  5. Enter the authentication token if the server requires one. The token is stored securely using encryption.
  6. Optionally add AI instructions that help the assistant understand when to use these tools (e.g., "Use these tools when a customer reports a bug or mentions an error message").
  7. Save the connection.

Testing a connection

After adding a connection, click the test button to verify it works. There There will connect to the MCP server, fetch the available tools, and report back with the number of tools found. If the test fails, check that the URL is correct and the authentication token is valid.

Successfully tested connections show a green status indicator and the number of available tools. The tool list is cached locally so the assistant does not need to fetch it on every conversation.

AI instructions

Each MCP connection has an optional AI instructions field. These instructions are included in the AI assistant's system prompt whenever the connection's tools are available. Use them to give the assistant context about what the tools do and when to use them.

For example:

"This connection provides access to our Flare error tracking project. When a customer mentions an error, exception, or bug, use these tools to look up recent occurrences and provide the customer with relevant details."

Good instructions help the assistant make better decisions about when to call external tools versus relying on its built-in knowledge.

Managing connections

You can edit a connection's name, URL, authentication token, and AI instructions at any time. Deleting a connection removes it and its cached tools immediately. The AI assistant will no longer have access to those tools in future conversations.

Use cases

Some examples of what MCP connections can enable:

  • Error tracking: Connect to Flare, Sentry, or Bugsnag so the assistant can look up errors by message or stack trace.
  • Project management: Connect to Linear or GitHub so the assistant can check issue status or link tickets to development work.
  • Billing and subscriptions: Connect to Stripe so the assistant can look up a customer's payment history or subscription status.
  • Internal APIs: Connect to any service that implements the MCP protocol to give the assistant access to your internal tools.