MCP

Expose tools that are hosted on an MCP server on the agentgateway.

About MCP

Model Context Protocol (MCP) is an open protocol that standardizes how Large Language Model (LLM) applications connect to various external data sources and tools. Without MCP, you need to implement custom integrations for each tool that your LLM application needs to access. However, this approach is hard to maintain and can cause issues when you want to scale your environment. With MCP, you can significantly speed up, simplify, and standardize these types of integrations.

An MCP server exposes external data sources and tools so that LLM applications can access them. Typically, you want to deploy these servers remotely and have authorization mechanisms in place so that LLM applications can safely access the data.

With agentgateway, you can connect to one or multiple MCP servers in any environment. The agentgateway proxies requests to the MCP tool that is exposed on the server. You can also use the agentgateway to federate tools from multiple MCP servers. For more information, see the MCP multiplexing guide.

Configure the agentgateway

1. Create an MCP configuration for your agentgateway.

   cat <<EOF > config.yaml
   binds:
      - port: 3000
        listeners:
        - routes:
          - policies:
              cors:
                allowOrigins:
                  - "*"
                allowHeaders:
                  - mcp-protocol-version
                  - content-type
            backends:
            - mcp:
                name: default
                targets:
                - name: everything
                  stdio:
                    cmd: npx
                    args: ["@modelcontextprotocol/server-everything"]
   EOF
   Review the following table to understand this configuration.

   Field	Description
   binds	Binds set up each port on which the agentgateway listens for incoming requests. This example configures one port.
   port	The port to listen on, such as 3000.
   listeners	Listeners configure how agentgateway accepts and processes incoming requests. They do this by grouping together resources such as routes and the backends that serve traffic.
   routes	Routes configure advanced routing features, such as traffic policies, that control how traffic is sent to the backends. In the example, the route matches all traffic.
   policies	Policies configure traffic policies on routes that shape how traffic is sent to the backends. In the example, a basic CORS policy is configured to allow all origins and the mcp-protocol-version header. This way, the configuration works with the MCP inspector tool.
   backends	Backends configure the backing destination where the traffic is sent to. Backends can be MCP servers, A2A agents, or OpenAPI servers. In this example, the backend is an MCP server.
   targets	Targets configure the details of the backend, such as the MCP server. In this example, the target is the sample, open source MCP test server, server-everything. The server runs a bunch of tools in a single process that are useful for testing.
   stdio	To run the server, you use the standard input/output (stdio) capability of the agentgateway, which allows you to pass in the command and command arguments that you want to use. In this example, the npx command is used. The npx command utility lets you run a Node.js package (@modelcontextprotocol/server-everything) without installing it. If you do not have npx on your machine, follow the instructions to install Node.js.

2. Run the agentgateway.

   agentgateway -f config.yaml