MCP Support in r-nacos

r-nacos supports MCP through a built-in MCP server and request forwarding. An ordinary HTTP endpoint registered with r-nacos can be exposed directly as an MCP service through r-nacos.

Use Cases

If one or more of your existing applications are already connected to the Nacos/r-nacos service registry and you want to expose MCP services to an AI coding agent or a custom agent, r-nacos requires no application code changes. Configure the corresponding MCP service in the console and it is ready to serve external clients.

If an application already provides a service through an MCP SDK but can run only as a single instance because it is stateful, r-nacos is also an option for enabling clustered deployment.

If the application is not yet registered with r-nacos, you can register it first and then use the r-nacos middleware service. r-nacos is compatible with the Nacos protocol and supports Nacos SDKs in many programming languages, making integration straightforward.

Features

1. Manage MCP tools and services in the console. MCP services support version management: the published and draft versions are maintained independently, and an earlier version can be restored quickly.
2. Support both the MCP SSE protocol and the MCP Streamable HTTP protocol (recommended).
3. Provide a built-in MCP service with request forwarding and several HTTP parameter mappings: pass the original JSON body, convert JSON to form parameters, or convert JSON to URL parameters.
4. Deliver high forwarding performance. In a single-node Streamable HTTP benchmark, throughput reached 29,350 QPS. SSE is stateful and currently lacks a suitable benchmarking tool; in theory, despite some additional intra-cluster forwarding, it should still exceed 15,000 QPS.
5. Support clustered calls. Multiple requests in one session may reach different r-nacos nodes; r-nacos uses its existing cluster communication layer to route each request to the node that owns the session.

Architecture

MCP architecture

Usage

1. Deploy and Run r-nacos

Use a v0.7.x release that supports MCP services. See the r-nacos deployment guide for details.

2. Create an MCP Tool Specification

A tool specification is the tool description presented to the large language model. It is independent of the concrete implementation, so one specification may have multiple implementations. A tool can also be associated with multiple MCP services, which is the main reason tool specifications are managed separately.

The following is a single-tool example for a large language model (OpenAI):

{
    "type": "function",
    "function": {
        "name": "add",
        "description": "add",
        "parameters": {
            "type": "object",
            "properties": {
                "a": {
                    "type": "number",
                    "description": "the first number"
                },
                "b": {
                    "type": "number",
                    "description": "the second number"
                }
            },
            "required": [
                "a",
                "b"
            ]
        }
    }
}

Ignoring the surrounding format, the specification contains three primary fields: name , description , and parameters . Together with the namespace and group required by r-nacos, a tool specification in r-nacos has five fields.

The details page is shown below:

MCP tool specification details

3. Create an MCP Service

An MCP service can be associated with a set of tools in the same namespace. To support tool calls, each tool requires both a specification and the corresponding call-routing information.

MCP service information:

MCP service information

Editing an individual tool:

MCP tool editor

Register the Service That Implements the Tool

For examples of a service used by the tool routes above, see the project's Python calculate-api or Rust calculate-api .

Before calling a tool, start the application targeted by the route and register it with r-nacos.

Use the Service

After creating the service, expose it through the following MCP endpoints:

Streamable HTTP:
http://{nacos_api_host}/rnacos/mcp/{server_unique_key}/{auth_key}
SSE:
http://{nacos_api_host}/rnacos/mcp/sse/{server_unique_key}/{auth_key}

Example:

Streamable: http://127.0.0.1:8848/rnacos/mcp/calculate/123456
SSE: http://127.0.0.1:8848/rnacos/mcp/sse/calculate/123456

Verify the Service

Run the official inspector with npx @modelcontextprotocol/inspector , then inspect and invoke the tools in your browser.

Verifying an MCP service

You can also configure the MCP service in Claude Code, Roo Code, OpenCode, another agent, or a custom agent.