examples/mcp/README.md
Examples demonstrating Model Context Protocol (MCP) integration with go-micro.
The simplest possible MCP-enabled service. Perfect for learning the basics.
What it shows:
Run it:
cd hello
go run main.go
A realistic service with create, read, update, delete, list, and search operations. Shows how to document a full API for agents with @example tags, description struct tags, validation errors, and partial updates.
Run it:
cd crud
go run main.go
Three services (Inventory, Orders, Notifications) showing how an AI agent orchestrates multi-step workflows: search products, check stock, reserve inventory, place order, send confirmation — all from a single natural language request.
Run it:
cd workflow
go run main.go
A complete platform (Users, Posts, Comments, Mail) mirroring micro/blog. Shows how existing microservices become agent-accessible with zero code changes — agents can sign up, write posts, comment, tag, and send mail through natural language.
Run it:
cd platform
go run main.go
Complete example showing all MCP features with a user service.
What it shows:
server.WithEndpointScopes()Run it:
cd documented
go run main.go
Add Go doc comments to your handler methods:
// SayHello greets a person by name. Returns a friendly greeting message.
//
// @example {"name": "Alice"}
func (g *Greeter) SayHello(ctx context.Context, req *HelloRequest, rsp *HelloResponse) error {
rsp.Message = "Hello " + req.Name + "!"
return nil
}
type HelloRequest struct {
Name string `json:"name" description:"Person's name to greet"`
}
handler := service.Server().NewHandler(new(Greeter))
service.Server().Handle(handler)
go mcp.ListenAndServe(":3000", mcp.Options{
Registry: service.Options().Registry,
})
# List tools
curl http://localhost:3000/mcp/tools | jq
# Call a tool
curl -X POST http://localhost:3000/mcp/call \
-H "Content-Type: application/json" \
-d '{
"tool": "greeter.Greeter.SayHello",
"input": {"name": "Alice"}
}' | jq
Start MCP server:
micro mcp serve
Add to ~/.claude/claude_desktop_config.json:
{
"mcpServers": {
"my-services": {
"command": "micro",
"args": ["mcp", "serve"]
}
}
}
Restart Claude Code and ask Claude to use your services!
Just write Go comments - documentation is extracted automatically:
# Start MCP server
micro mcp serve # Stdio (for Claude Code)
micro mcp serve --address :3000 # HTTP/SSE (for web agents)
# List available tools
micro mcp list # Human-readable list
micro mcp list --json # JSON output
# Test a tool
micro mcp test <tool-name> '{"key": "value"}'
# Generate documentation
micro mcp docs # Markdown format
micro mcp docs --format json # JSON format
micro mcp docs --output tools.md # Save to file
# Export to different formats
micro mcp export langchain # Python LangChain tools
micro mcp export openapi # OpenAPI 3.0 spec
micro mcp export json # Raw JSON definitions
For detailed examples, see CLI Examples.
Declare required scopes when registering a handler:
handler := service.Server().NewHandler(
new(BlogService),
server.WithEndpointScopes("Blog.Create", "blog:write"),
server.WithEndpointScopes("Blog.Delete", "blog:admin"),
)
Or define scopes at the gateway layer without changing services:
mcp.Serve(mcp.Options{
Registry: reg,
Auth: authProvider,
Scopes: map[string][]string{
"blog.Blog.Create": {"blog:write"},
"blog.Blog.Delete": {"blog:admin"},
},
})
Every tool call generates a trace ID that propagates through the RPC chain. Configure rate limiting and audit logging at the gateway:
mcp.Serve(mcp.Options{
Registry: reg,
Auth: authProvider,
RateLimit: &mcp.RateLimitConfig{
RequestsPerSecond: 10,
Burst: 20,
},
AuditFunc: func(r mcp.AuditRecord) {
log.Printf("[audit] trace=%s tool=%s account=%s allowed=%v",
r.TraceID, r.Tool, r.AccountID, r.Allowed)
},
})