Fundamentals
What is MCP and why it matters for customer support
By the RoutMe team. Last updated .
The Model Context Protocol (MCP) is an open standard that lets AI applications discover and call tools exposed by other systems. For customer support, it means an assistant can look up an order or an account through a defined interface on your own systems, instead of guessing or relying on a copied export.
What MCP is
MCP defines how a client application, which hosts the model, talks to servers that expose capabilities. The specification describes tools as functions a language model can invoke, each identified by a name and described with metadata and a JSON schema for its inputs. The client lists the tools with a tools/list request and invokes one with tools/call. The result comes back as content the model can read.
That is all that is needed for support. A business exposes a handful of tools: "get order", "search products", "get account plan". The assistant picks one based on the question, calls it, and writes an answer from the result.
Why it matters for customer support
| Without a tool interface | With MCP tools |
|---|---|
| Answers from static documents that go stale | Answers from the current record |
| Staff paste statuses into replies | The assistant reads the status itself |
| A custom integration per assistant | One tool server usable by any MCP client |
| Broad database access or none | Narrow, named operations |
The last row matters most. You do not hand the assistant your database. You hand it a few operations, each with a clear input and output, that you can review the way you review any API.
Safety belongs to the design
The MCP tools specification says there should always be a human in the loop with the ability to deny tool invocations in applications that expose tools to a model, and that servers must validate inputs, apply access controls, rate limit and sanitize outputs. It also warns that clients must consider tool annotations untrusted unless they come from trusted servers.
So an annotation such as readOnlyHint: true is a statement by the server author, useful when you control the server. RoutMe's chat assistant is only offered tools whose server marks them read-only, and flow Agent nodes use the servers you attach to the node, so only attach servers you control. For how to build the boundary, see what read-only AI means.
How RoutMe uses MCP
In RoutMe you register a server by URL, with an optional bearer token, test the connection, and choose which widgets may call it. RoutMe speaks MCP's Streamable HTTP transport and calls initialize, tools/list and tools/call, with a 20 second timeout per request. If one server is down, RoutMe skips it and carries on with the others. Read the MCP documentation.
Connecting a store follows the same pattern on any platform: your MCP server calls your platform's API with credentials you hold. See Shopify and other platforms. Turning a REST or OpenAPI description into MCP tools with one import is in progress and listed as coming soon.
What MCP does not do
- It does not decide what the model may do. Your tool list does.
- It does not verify the customer. Your tool must check ownership, for example order number plus email.
- It does not make answers correct. Logging and handoff still matter.
Designing good support tools
Tool design decides how well an assistant behaves. Keep each tool narrow and named for the question it answers, such as get_order_status, rather than a generic query tool. Describe the tool and each parameter clearly, because the model reads those descriptions when it chooses what to call. Return only the fields needed to answer, in a compact form, and return a plain error message when nothing matches. Keep tools fast, since the customer is waiting: RoutMe applies a 20 second timeout per request, and a single assistant turn allows up to six rounds of tool calls. Finally, test each tool with Test connection in the dashboard to confirm it is listed.
Sources
Explore the product: RoutMe for Shopify, solutions and the documentation.