go-mcp
Categories
Language:
Go
Stars:
24
Forks:
2
go-mcp
A Go implementation of the Model Context Protocol (MCP) - an open protocol that enables seamless integration between LLM applications and external data sources and tools.
⚠️ Warning: The main branch contains unreleased changes and may be unstable. We recommend using the latest tagged release for stability. This library follows semantic versioning - breaking changes may be introduced with minor version bumps (0.x.0) until v1.0.0 is released. After v1.0.0, the API will be stable and breaking changes will only occur in major version updates. We recommend pinning your dependency to a specific version and reviewing the changelog before upgrading.
Overview
This repository provides a Go library implementing the Model Context Protocol (MCP) following the official specification.
Features
Core Protocol
- Complete MCP protocol implementation with JSON-RPC 2.0 messaging
- Pluggable transport system supporting SSE and Standard IO
- Session-based client-server communication
- Comprehensive error handling and progress tracking
Server Features
- Modular server implementation with optional capabilities
- Support for prompts, resources, and tools
- Real-time notifications and updates
- Built-in logging system
- Resource subscription management
Client Features
-
Flexible client configuration with optional capabilities
-
Automatic session management and health monitoring
- Support for streaming and pagination
-
Progress tracking and cancellation support
-
Configurable timeouts and retry logic
Transport Options
- Server-Sent Events (SSE) for web-based real-time updates
- Standard IO for command-line tool integration
Installation
go get github.com/MegaGrindStone/go-mcp
Usage
Server Implementation
There are two main steps to implementing an MCP server:
1. Create a Server Implementation
Create a server implementation that provides the capabilities you need:
// Example implementing a server with tool support
type MyToolServer struct{}
func (s *MyToolServer) ListTools(ctx context.Context, params mcp.ListToolsParams,
progress mcp.ProgressReporter, requestClient mcp.RequestClientFunc) (mcp.ListToolsResult, error) {
// Return available tools
return mcp.ListToolsResult{
Tools: []mcp.Tool{
{
Name: "example-tool",
Description: "An example tool",
// Additional tool properties...
},
},
}, nil
}
func (s *MyToolServer) CallTool(ctx context.Context, params mcp.CallToolParams,
progress mcp.ProgressReporter, requestClient mcp.RequestClientFunc) (mcp.CallToolResult, error) {
// Implement tool functionality
return mcp.CallToolResult{
Content: []mcp.Content{
{
Type: mcp.ContentTypeText,
Text: "Tool result",
},
},
}, nil
}
2. Initialize and Serve
Create and configure the server with your implementation and chosen transport:
// Create server with your implementation
toolServer := &MyToolServer{}
// Choose a transport method
// Option 1: Server-Sent Events (SSE)
sseSrv := mcp.NewSSEServer("/message")
srv := mcp.NewServer(mcp.Info{
Name: "my-mcp-server",
Version: "1.0",
}, sseSrv,
mcp.WithToolServer(toolServer),
mcp.WithServerPingInterval(30*time.Second),
// Add other capabilities as needed
)
// Set up HTTP handlers for SSE
http.Handle("/sse", sseSrv.HandleSSE())
http.Handle("/message", sseSrv.HandleMessage())
go http.ListenAndServe(":8080", nil)
// Option 2: Standard IO
srvIO := mcp.NewStdIO(os.Stdin, os.Stdout)
srv := mcp.NewServer(mcp.Info{
Name: "my-mcp-server",
Version: "1.0",
}, srvIO,
mcp.WithToolServer(toolServer),
// Add other capabilities as needed
)
// Start the server - this blocks until shutdown
go srv.Serve()
// To shutdown gracefully
ctx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
defer cancel()
srv.Shutdown(ctx)
Available Server Options
Configure your server with additional capabilities:
// Prompt capabilities
mcp.WithPromptServer(promptServer)
mcp.WithPromptListUpdater(promptListUpdater)
// Resource capabilities
mcp.WithResourceServer(resourceServer)
mcp.WithResourceListUpdater(resourceListUpdater)
mcp.WithResourceSubscriptionHandler(subscriptionHandler)
// Tool capabilities
mcp.WithToolServer(toolServer)
mcp.WithToolListUpdater(toolListUpdater)
// Roots and logging capabilities
mcp.WithRootsListWatcher(rootsListWatcher)
mcp.WithLogHandler(logHandler)
// Server behavior configuration
mcp.WithServerPingInterval(interval)
mcp.WithServerPingTimeout(timeout)
mcp.WithServerPingTimeoutThreshold(threshold)
mcp.WithServerSendTimeout(timeout)
mcp.WithInstructions(instructions)
// Event callbacks
mcp.WithServerOnClientConnected(func(id string, info mcp.Info) {
fmt.Printf("Client connected: %s\n", id)
})
mcp.WithServerOnClientDisconnected(func(id string) {
fmt.Printf("Client disconnected: %s\n", id)
})
Client Implementation
The client implementation involves creating a client with transport options and capabilities, connecting to a server, and executing MCP operations.
Creating and Connecting a Client
// Create client info
info := mcp.Info{
Name: "my-mcp-client",
Version: "1.0",
}
// Create a context for connection and operations
ctx, cancel := context.WithCancel(context.Background())
defer cancel()
// Choose transport layer - SSE or Standard IO
// Option 1: Server-Sent Events (SSE)
sseClient := mcp.NewSSEClient("http://localhost:8080/sse", http.DefaultClient)
cli := mcp.NewClient(info, sseClient,
// Optional client configurations
mcp.WithClientPingInterval(30*time.Second),
mcp.WithProgressListener(progressListener),
mcp.WithLogReceiver(logReceiver),
)
// Option 2: Standard IO
srvReader, srvWriter := io.Pipe()
cliReader, cliWriter := io.Pipe()
cliIO := mcp.NewStdIO(cliReader, srvWriter)
srvIO := mcp.NewStdIO(srvReader, cliWriter)
cli := mcp.NewClient(info, cliIO)
// Connect client (requires context)
if err := cli.Connect(ctx); err != nil {
log.Fatal(err)
}
// Ensure proper cleanup
defer func() {
disconnectCtx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
defer cancel()
cli.Disconnect(disconnectCtx)
}()
Making Requests
// List available tools
tools, err := cli.ListTools(ctx, mcp.ListToolsParams{})
if err != nil {
log.Fatal(err)
}
// Call a tool (with proper argument structure)
args := map[string]string{"message": "Hello MCP!"}
argsBs, _ := json.Marshal(args)
result, err := cli.CallTool(ctx, mcp.CallToolParams{
Name: "echo",
Arguments: argsBs,
})
if err != nil {
log.Fatal(err)
}
// Work with resources
resources, err := cli.ListResources(ctx, mcp.ListResourcesParams{})
if err != nil {
log.Fatal(err)
}
// Subscribe to resource updates
err = cli.SubscribeResource(ctx, mcp.SubscribeResourceParams{
URI: "resource-uri",
})
if err != nil {
log.Fatal(err)
}
// Work with prompts
prompts, err := cli.ListPrompts(ctx, mcp.ListPromptsParams{})
if err != nil {
log.Fatal(err)
}
prompt, err := cli.GetPrompt(ctx, mcp.GetPromptParams{
Name: "my-prompt",
})
if err != nil {
log.Fatal(err)
}
#### Implementing Handlers for Client Capabilities
// Implement required interfaces for client capabilities
type myClient struct {
// ...client fields
}
// For sampling capability
func (c *myClient) CreateSampleMessage(ctx context.Context, params mcp.SamplingParams) (mcp.SamplingResult, error) {
// Generate sample LLM output
return mcp.SamplingResult{
Role: mcp.RoleAssistant,
Content: mcp.SamplingContent{
Type: mcp.ContentTypeText,
Text: "Sample response text",
},
Model: "my-llm-model",
}, nil
}
// For resource subscription notifications
func (c *myClient) OnResourceSubscribedChanged(uri string) {
fmt.Printf("Resource %s was updated\n", uri)
}
// For progress tracking
func (c *myClient) OnProgress(params mcp.ProgressParams) {
fmt.Printf("Progress: %.2f/%.2f\n", params.Progress, params.Total)
}
// Pass these handlers when creating the client
cli := mcp.NewClient(info, transport,
mcp.WithSamplingHandler(client),
mcp.WithResourceSubscribedWatcher(client),
mcp.WithProgressListener(client),
)
Complete Examples
For complete working examples:
- See
example/everything/
for a comprehensive server and client implementation with all features - See
example/filesystem/
for a focused example of file operations using Standard IO transport
These examples demonstrate:
- Server and client lifecycle management
- Transport layer setup
- Error handling
- Tool implementation
- Resource management
- Progress tracking
- Logging integration
For more details, check the example directory in the repository.
Server Packages
The servers
directory contains reference server implementations that mirror those found in the official modelcontextprotocol/servers repository.
Contributing
Contributions are welcome! Please feel free to submit a Pull Request. For major changes, please open an issue first to discuss what you would like to change.
License
This project is licensed under the MIT License - see the LICENSE file for details.
Publisher info
More MCP servers built with Go
A proof-of-concept demonstrating a custom-built host implementing an OpenAI-compatible API with Google Gemini, function calling, and interaction with a dummy MCP server.