mcpField
G
grpcmcp

grpcmcp

Local
github.com/adiom-dataDeveloper Tools29Updated 21w ago

About grpcmcp

grpcmcp is an MCP server that proxies requests to a gRPC backend, using either a protobuf descriptor file or gRPC reflection. It suits developers who want AI assistants to call existing gRPC services.

Key features

  1. 01Exposes unary gRPC methods as MCP tools
  2. 02Loads schemas from descriptor files or reflection
  3. 03Supports Streamable HTTP, SSE and STDIN transports
  4. 04Filters services by fully qualified name
  5. 05Supports bearer tokens and custom headers
  6. 06Can output 64-bit integers as strings

Use cases

  1. 01Letting an AI assistant call internal gRPC microservices
  2. 02Testing gRPC interfaces during development
  3. 03Embedding a gRPC-to-MCP handler in a Go application

MCP support

grpcmcp is an MCP server that presents unary methods of a backend gRPC service as MCP tools to MCP clients. Client-streaming, server-streaming and bidi-streaming methods are skipped.

Configuration

Local server config is the same for every client. Copy it into your client’s MCP config file.

No install method listed yet

We haven’t collected a package or remote endpoint for this server. Follow the instructions in the project README.

Repository

Tools

No tool list yet

Local servers expose tools only once running on your machine, and the maintainer has not provided a tool list.

Details

From the project README

grpcmcp

A simple MCP server that will proxy to a grpc backend based on a provided descriptors file or using reflection.

Quick Start

  1. Install the binary: go install . or go install github.com/adiom-data/grpcmcp Ensure the go bin directory is in your PATH.

  2. In a terminal, run the example grpc server go run example/main.go. This will start a grpc health service on port 8090 with server reflection enabled. Note that this runs on the default port that grpcmcp will connect to.

  3. Streamable HTTP Transport In another terminal, run grpcmcp --hostport=localhost:3000 --reflect. Specifying hostport will use Streamable HTTP by default. The MCP endpoint will be served at http://localhost:3000/mcp.

  4. Legacy SSE Transport For older clients, run grpcmcp --hostport=localhost:3000 --transport=sse --reflect. The SSE endpoint will be served at http://localhost:3000/sse.

  5. STDIN Transport Set up the MCP config. e.g.

"grpcmcp": {
    "command": "grpcmcp",
    "args": ["--reflect"]
}

Options / Features

grpcmcp --help for a full list of options.

  • hostport string - When set, serve MCP over HTTP, and use this as the server host:port.

  • transport string - Transport to use when hostport is set. Defaults to http for Streamable HTTP at /mcp. Set to sse for the legacy SSE transport at /sse.

  • descriptors string - Specify file location of the protobuf definitions generated from buf build -o protos.pb or protoc --descriptor_set_out=protos.pb instead of using gRPC reflection.

  • reflect - If set, use reflection to retrieve gRPC endpoints instead of descriptor file.

  • url string - Specify the url of the backend server.

  • services string - Comma separated list of fully qualified gRPC service names to filter.

  • bearer string - Token to attach in an Authorization: Bearer header.

  • bearer-env string - Environment variable for token to attach in an Authorization: Bearer header. Overrides bearer.

  • header string (repeatable) - Headers to add in Key: Value format.

  • string64 - If set, expose 64-bit protobuf integer fields (int64, uint64, sint64, fixed64, sfixed64) as strings only in MCP JSON schemas. This avoids precision ambiguity for JavaScript-based clients and agents. By default, schemas continue to allow either JSON numbers or strings for compatibility.

grpcmcp currently exposes unary gRPC methods as MCP tools. Backend gRPC client-streaming, server-streaming, and bidi-streaming methods are skipped. This is separate from MCP transports: Streamable HTTP and legacy SSE are supported for client connections to grpcmcp.

Library Usage

grpcmcp can also be embedded as a Go library. The library accepts a descriptor set directly, so applications can decide how to load descriptors and how to wrap the MCP HTTP handler.

descriptors, err := grpcmcp.LoadDescriptorsFromReflection(ctx, backendURL, headers, false)
if err != nil {
    return err
}

srv, err := grpcmcp.NewServer(grpcmcp.Config{
    ServerName:  "gRPC MCP Server",
    Version:     "1.0.0",
    BaseURL:     backendURL,
    Descriptors: descriptors,
    Headers:     grpcmcp.StaticHeaders(headers),
})
if err != nil {
    return err
}

handler := server.NewStreamableHTTPServer(srv)

To route backend calls through a custom transport, provide HTTPClient. For example, an embedded application can use an in-memory HTTP transport such as go.akshayshah.org/memhttp by passing the in-memory server's URL and client:

backend, err := memhttp.New(backendHandler)
if err != nil {
    return err
}
defer backend.Close()

srv, err := grpcmcp.NewServer(grpcmcp.Config{
    BaseURL:     backend.URL(),
    HTTPClient:  backend.Client(),
    Descriptors: descriptors,
})

For dynamic backend auth, provide a ToolHeaderProvider. The full mcp.CallToolRequest is available, including inbound HTTP headers supplied by supported MCP transports.

srv, err := grpcmcp.NewServer(grpcmcp.Config{
    BaseURL:     backendURL,
    Descriptors: descriptors,
    Headers: func(ctx context.Context, req mcp.CallToolRequest) (http.Header, error) {
        h := make(http.Header)
        h.Set("Authorization", req.Header.Get("Authorization"))
        return h, nil
    },
})

Inbound MCP authentication should be handled by wrapping the HTTP handler with standard Go middleware. Outbound backend authentication is controlled by the configured header provider.

Help

Join our Discord at https://discord.gg/hDjx3DehwG

Comments

Markdown supported

New to MCP?

From what the protocol is to installing your first server, these answers cover most first questions.

An open protocol that defines how AI applications connect to external tools and data in a uniform way. Clients talk to the model; MCP servers provide tools, resources and prompt templates. They exchange standard messages, so one server can be reused by many clients.