CallToolParams.Arguments is any. Clients can send whatever they want. Tool authors who create their own ToolHandler (bypassing NewTool) will receive the arguments as a map[string]any which has been validated against the input schema. Tool authors who call NewTool can choose the type that the arguments unmarshal into. There is no double-unmarshaling and no downcasting. There is no way for tool authors to avoid the unmarshal, and no way to avoid the validation unless the omit an input schema, which is probably bad for LLMs. If tool authors want this optimization, we can provide it later. Change-Id: Ieb201e56c7fb8a23bb1a93b27e946630ec9e79ad Reviewed-on: https://go-review.googlesource.com/c/tools/+/679595 TryBot-Bypass: Jonathan Amsterdam <jba@google.com> Auto-Submit: Jonathan Amsterdam <jba@google.com> Reviewed-by: Robert Findley <rfindley@google.com>
3.5 KiB
MCP SDK prototype
Contents
The mcp package provides a software development kit (SDK) for writing clients and servers of the model context protocol. It is unstable, and will change in breaking ways in the future. As of writing, it is a prototype to explore the design space of client/server transport and binding.
Installation
The mcp package is currently internal and cannot be imported using go get.
Quickstart
Here's an example that creates a client that talks to an MCP server running as a sidecar process:
package main
import (
"context"
"log"
"os/exec"
"github.com/modelcontextprotocol/go-sdk/mcp"
)
func main() {
ctx := context.Background()
// Create a new client, with no features.
client := mcp.NewClient("mcp-client", "v1.0.0", nil)
// Connect to a server over stdin/stdout
transport := mcp.NewCommandTransport(exec.Command("myserver"))
session, err := client.Connect(ctx, transport)
if err != nil {
log.Fatal(err)
}
defer session.Close()
// Call a tool on the server.
params := &mcp.CallToolParams{
Name: "greet",
Arguments: map[string]any{"name": "you"},
}
if res, err := session.CallTool(ctx, params); err != nil {
log.Printf("CallTool failed: %v", err)
} else {
if res.IsError {
log.Print("tool failed")
}
for _, c := range res.Content {
log.Print(c.Text)
}
}
}
Here is an example of the corresponding server, connected over stdin/stdout:
package main
import (
"context"
"github.com/modelcontextprotocol/go-sdk/mcp"
)
type HiParams struct {
Name string `json:"name"`
}
func SayHi(ctx context.Context, cc *mcp.ServerSession, params *mcp.CallToolParamsFor[HiParams]) (*mcp.CallToolResultFor[string], error) {
return &mcp.CallToolResultFor[string]{
Content: []*mcp.Content{mcp.NewTextContent("Hi " + params.Name)},
}, nil
}
func main() {
// Create a server with a single tool.
server := mcp.NewServer("greeter", "v1.0.0", nil)
server.AddTools(mcp.NewTool("greet", "say hi", SayHi))
// Run the server over stdin/stdout, until the client disconnects
_ = server.Run(context.Background(), mcp.NewStdIOTransport())
}
Design
See design.md for the SDK design. That document is canonical: given any divergence between the design doc and this prototype, the doc reflects the latest design.
Testing
To test your client or server using stdio transport, you can use an in-memory transport. See example.
To test your client or server using sse transport, you can use the httptest package. See example.
Code of Conduct
This project follows the Go Community Code of Conduct. If you encounter a conduct-related issue, please mail conduct@golang.org.
License
Unless otherwise noted, the Go source files are distributed under the BSD-style license found in the LICENSE file.
Upon a potential move to the modelcontextprotocol organization, the license will be updated to the MIT License, and the license header will reflect the Go MCP SDK Authors.