Tyk AI Studio Plugin Best Practices

Availability

Edition	Deployment Type
Community & Enterprise	Self-Managed, Hybrid

Production-ready patterns, performance optimization, and security guidelines for Tyk AI Studio plugins using the Unified Plugin SDK.

Architecture

Single Responsibility

Each plugin should have a clear, focused purpose:

// Good: Focused on one concern
type RateLimiterPlugin struct {
    plugin_sdk.BasePlugin
}

// Avoid: Doing too many unrelated things
type EverythingPlugin struct {
    plugin_sdk.BasePlugin
    // rate limiting + auth + logging + analytics + ...
}

When to combine capabilities:

✅ UI + PostAuth for rate limiting dashboard
✅ PostAuth + Response for request/response correlation
✅ Object Hooks + UI for approval workflows
❌ Unrelated features that could be separate plugins

BasePlugin Usage

Always use BasePlugin for consistent lifecycle management:

Expandable

type MyPlugin struct {
    plugin_sdk.BasePlugin
    config *Config
    client *http.Client
}

func NewMyPlugin() *MyPlugin {
    return &MyPlugin{
        BasePlugin: plugin_sdk.NewBasePlugin(
            "my-plugin",
            "1.0.0",
            "Clear description of what it does",
        ),
    }
}

Configuration Management

Parse configuration in Initialize(), validate early:

Expandable

func (p *MyPlugin) Initialize(ctx plugin_sdk.Context, config map[string]string) error {
    // Extract broker ID for Service API
    if brokerIDStr, ok := config["_service_broker_id"]; ok {
        var brokerID uint32
        fmt.Sscanf(brokerIDStr, "%d", &brokerID)
        ai_studio_sdk.SetServiceBrokerID(brokerID)
    }

    // Validate required configuration
    apiKey, ok := config["api_key"]
    if !ok || apiKey == "" {
        return fmt.Errorf("api_key is required")
    }

    // Parse optional configuration with defaults
    timeout := 30
    if timeoutStr, ok := config["timeout"]; ok {
        if t, err := strconv.Atoi(timeoutStr); err == nil {
            timeout = t
        }
    }

    // Initialize resources
    p.config = &Config{
        APIKey:  apiKey,
        Timeout: time.Duration(timeout) * time.Second,
    }

    p.client = &http.Client{
        Timeout: p.config.Timeout,
    }

    ctx.Services.Logger().Info("Plugin initialized",
        "timeout", timeout,
    )

    return nil
}

Error Handling

Fail Fast, Fail Clearly

Return errors early with descriptive messages:

Expandable

func (p *MyPlugin) HandlePostAuth(ctx plugin_sdk.Context, req *pb.EnrichedRequest) (*pb.PluginResponse, error) {
    // Validate input
    if req.Method != "POST" {
        return &pb.PluginResponse{
            Block:        true,
            ErrorMessage: "Only POST requests are allowed",
        }, nil
    }

    // Check external dependency
    status, err := p.checkExternalService(ctx)
    if err != nil {
        ctx.Services.Logger().Error("External service check failed",
            "error", err,
            "app_id", ctx.AppID,
        )
        return &pb.PluginResponse{
            Block:        true,
            ErrorMessage: "Service temporarily unavailable",
        }, nil
    }

    return &pb.PluginResponse{Modified: false}, nil
}

Log Errors with Context

Always include relevant context in error logs:

ctx.Services.Logger().Error("Failed to validate request",
    "error", err,
    "app_id", ctx.AppID,
    "user_id", ctx.UserID,
    "path", req.Path,
    "method", req.Method,
)

Graceful Degradation

Don’t fail hard if non-critical operations fail:

// Bad: Plugin fails if KV write fails
err := ctx.Services.KV().Write(ctx, "cache", data)
if err != nil {
    return nil, err // Blocks entire request!
}

// Good: Log and continue
err := ctx.Services.KV().Write(ctx, "cache", data)
if err != nil {
    ctx.Services.Logger().Warn("Failed to cache data", "error", err)
    // Continue processing
}

Performance

Minimize Blocking Operations

Avoid blocking in request path:

Expandable

// Bad: Synchronous external call in request path
func (p *MyPlugin) HandlePostAuth(ctx plugin_sdk.Context, req *pb.EnrichedRequest) (*pb.PluginResponse, error) {
    // This blocks the request!
    result, err := p.callSlowExternalAPI(req.Body)
    return &pb.PluginResponse{Modified: false}, nil
}

// Good: Async processing for non-critical operations
func (p *MyPlugin) HandlePostAuth(ctx plugin_sdk.Context, req *pb.EnrichedRequest) (*pb.PluginResponse, error) {
    // Fire and forget for analytics
    go func() {
        p.trackRequest(req)
    }()

    return &pb.PluginResponse{Modified: false}, nil
}

Use Connection Pooling

Reuse HTTP clients and database connections. The SDK provides a DefaultHTTPClient() with sensible defaults (30s timeout, connection pooling, TLS handshake timeout):

func NewMyPlugin() *MyPlugin {
    return &MyPlugin{
        BasePlugin: plugin_sdk.NewBasePlugin("my-plugin", "1.0.0", "desc"),
        // Use the SDK default client with connection pooling
        httpClient: plugin_sdk.DefaultHTTPClient(),
    }
}

You can also configure your own client if you need different settings:

httpClient: &http.Client{
    Timeout: 10 * time.Second,
    Transport: &http.Transport{
        MaxIdleConns:        100,
        MaxIdleConnsPerHost: 10,
        IdleConnTimeout:     90 * time.Second,
    },
}

Important: Always use http.NewRequestWithContext(ctx, ...) to respect context cancellation and timeouts. Never use the default http.Client{} with zero timeout in production plugins.

Cache Frequently Accessed Data

Use KV storage or in-memory caching:

Expandable

type MyPlugin struct {
    plugin_sdk.BasePlugin
    cache     sync.Map // In-memory cache
    cacheTime time.Duration
}

func (p *MyPlugin) getConfig(ctx plugin_sdk.Context, appID uint32) (*AppConfig, error) {
    // Check memory cache first
    key := fmt.Sprintf("config:%d", appID)
    if val, ok := p.cache.Load(key); ok {
        cached := val.(*cachedConfig)
        if time.Since(cached.timestamp) < p.cacheTime {
            return cached.config, nil
        }
    }

    // Check KV storage
    data, err := ctx.Services.KV().Read(ctx, key)
    if err == nil {
        var config AppConfig
        json.Unmarshal(data, &config)

        // Update memory cache
        p.cache.Store(key, &cachedConfig{
            config:    &config,
            timestamp: time.Now(),
        })

        return &config, nil
    }

    // Fetch from Service API (slowest)
    app, err := ctx.Services.Studio().GetApp(ctx, appID)
    // ... cache and return
}

Batch Operations

Batch external calls when possible:

// Bad: N+1 queries
for _, appID := range appIDs {
    app, _ := ctx.Services.Studio().GetApp(ctx, appID)
    // process app
}

// Good: Batch fetch
apps, _ := ctx.Services.Studio().ListApps(ctx, 1, 100)

Security

Input Validation

Always validate and sanitize inputs:

Expandable

func (p *MyPlugin) HandleObjectHook(ctx plugin_sdk.Context, req *pb.ObjectHookRequest) (*pb.ObjectHookResponse, error) {
    // Validate object JSON
    var llm LLM
    if err := json.Unmarshal([]byte(req.ObjectJson), &llm); err != nil {
        return &pb.ObjectHookResponse{
            AllowOperation:  false,
            RejectionReason: "Invalid LLM configuration",
        }, nil
    }

    // Validate required fields
    if llm.Name == "" {
        return &pb.ObjectHookResponse{
            AllowOperation:  false,
            RejectionReason: "LLM name is required",
        }, nil
    }

    // Validate URL format
    if llm.APIEndpoint != "" {
        u, err := url.Parse(llm.APIEndpoint)
        if err != nil || (u.Scheme != "https" && u.Scheme != "http") {
            return &pb.ObjectHookResponse{
                AllowOperation:  false,
                RejectionReason: "Invalid API endpoint URL",
            }, nil
        }
    }

    return &pb.ObjectHookResponse{AllowOperation: true}, nil
}

Least Privilege Permissions

Only request permissions you actually need:

Expandable

// Bad: Requesting everything
{
  "permissions": {
    "services": [
      "llms.read", "llms.write", "llms.proxy",
      "tools.read", "tools.write", "tools.execute",
      "apps.read", "apps.write",
      "analytics.read", "kv.readwrite"
    ]
  }
}

// Good: Only what's needed
{
  "permissions": {
    "services": [
      "llms.read",      // Read LLM configs
      "kv.readwrite"    // Store plugin state
    ]
  }
}

Secrets Management

Never hardcode secrets, use configuration:

// Bad: Hardcoded secrets
apiKey := "sk-1234567890abcdef"

// Good: From configuration
apiKey, ok := config["api_key"]
if !ok {
    return fmt.Errorf("api_key configuration required")
}

// Better: From environment or secrets manager
apiKey := os.Getenv("PLUGIN_API_KEY")

Sanitize Logs

Don’t log sensitive data:

// Bad: Logging sensitive data
ctx.Services.Logger().Info("Request received",
    "headers", req.Headers, // May contain auth tokens!
    "body", req.Body,       // May contain PII!
)

// Good: Log only safe metadata
ctx.Services.Logger().Info("Request received",
    "app_id", ctx.AppID,
    "method", req.Method,
    "path", req.Path,
    "content_length", len(req.Body),
)

Observability

Structured Logging

Use key-value pairs for searchable logs:

Expandable

// Good structured logging
ctx.Services.Logger().Info("Request processed",
    "app_id", ctx.AppID,
    "user_id", ctx.UserID,
    "duration_ms", time.Since(startTime).Milliseconds(),
    "status", "success",
)

ctx.Services.Logger().Error("External API call failed",
    "error", err,
    "endpoint", endpoint,
    "status_code", statusCode,
    "retry_count", retryCount,
)

Request Tracing

Include request IDs for correlation:

Expandable

func (p *MyPlugin) HandlePostAuth(ctx plugin_sdk.Context, req *pb.EnrichedRequest) (*pb.PluginResponse, error) {
    requestID := generateRequestID()

    // Add to request for downstream
    if req.Headers == nil {
        req.Headers = make(map[string]string)
    }
    req.Headers["X-Request-ID"] = requestID

    ctx.Services.Logger().Info("Processing request",
        "request_id", requestID,
        "app_id", ctx.AppID,
    )

    // Use request ID in all logs for this request
    defer func() {
        ctx.Services.Logger().Info("Request completed",
            "request_id", requestID,
            "duration_ms", time.Since(startTime).Milliseconds(),
        )
    }()

    return &pb.PluginResponse{
        Modified: true,
        Request:  req,
    }, nil
}

Metrics Collection

Track plugin performance:

Expandable

type MyPlugin struct {
    plugin_sdk.BasePlugin
    stats struct {
        sync.RWMutex
        requestCount   int64
        errorCount     int64
        totalDuration  time.Duration
    }
}

func (p *MyPlugin) HandlePostAuth(ctx plugin_sdk.Context, req *pb.EnrichedRequest) (*pb.PluginResponse, error) {
    startTime := time.Now()
    defer func() {
        duration := time.Since(startTime)

        p.stats.Lock()
        p.stats.requestCount++
        p.stats.totalDuration += duration
        p.stats.Unlock()

        if duration > 100*time.Millisecond {
            ctx.Services.Logger().Warn("Slow plugin execution",
                "duration_ms", duration.Milliseconds(),
            )
        }
    }()

    // ... plugin logic
}

// Expose metrics via RPC (for UI plugins)
func (p *MyPlugin) HandleCall(method string, payload []byte) ([]byte, error) {
    if method == "get_metrics" {
        p.stats.RLock()
        defer p.stats.RUnlock()

        return json.Marshal(map[string]interface{}{
            "request_count":   p.stats.requestCount,
            "error_count":     p.stats.errorCount,
            "avg_duration_ms": p.stats.totalDuration.Milliseconds() / p.stats.requestCount,
        })
    }
    return nil, fmt.Errorf("unknown method")
}

Testing

Unit Tests

Test plugin logic in isolation:

Expandable

func TestValidation(t *testing.T) {
    plugin := NewValidatorPlugin()

    tests := []struct {
        name        string
        llm         LLM
        shouldAllow bool
        reason      string
    }{
        {
            name:        "valid HTTPS endpoint",
            llm:         LLM{APIEndpoint: "https://api.example.com"},
            shouldAllow: true,
        },
        {
            name:        "HTTP endpoint blocked",
            llm:         LLM{APIEndpoint: "http://api.example.com"},
            shouldAllow: false,
            reason:      "API endpoint must use HTTPS",
        },
    }

    for _, tt := range tests {
        t.Run(tt.name, func(t *testing.T) {
            llmJSON, _ := json.Marshal(tt.llm)
            req := &pb.ObjectHookRequest{
                ObjectType: "llm",
                ObjectJson: string(llmJSON),
            }

            resp, err := plugin.HandleObjectHook(mockContext(), req)
            assert.NoError(t, err)
            assert.Equal(t, tt.shouldAllow, resp.AllowOperation)
            if !tt.shouldAllow {
                assert.Contains(t, resp.RejectionReason, tt.reason)
            }
        })
    }
}

Integration Tests

Test with Service API:

Expandable

func TestServiceAPIIntegration(t *testing.T) {
    // Requires test environment
    if os.Getenv("INTEGRATION_TEST") == "" {
        t.Skip("Skipping integration test")
    }

    plugin := NewMyPlugin()
    ctx := testContext()

    // Test KV operations
    err := ctx.Services.KV().Write(ctx, "test-key", []byte("value"))
    assert.NoError(t, err)

    data, err := ctx.Services.KV().Read(ctx, "test-key")
    assert.NoError(t, err)
    assert.Equal(t, "value", string(data))
}

Object Hooks Best Practices

Hook Priority

Use priority to control execution order:

0-10: Critical validation (security, compliance)
11-50: Business logic validation
51-100: Enrichment and metadata

func (p *SecurityPlugin) GetObjectHookRegistrations() ([]*pb.ObjectHookRegistration, error) {
    return []*pb.ObjectHookRegistration{
        {
            ObjectType: "llm",
            HookTypes:  []string{"before_create", "before_update"},
            Priority:   5, // Run early for security checks
        },
    }, nil
}

Before vs After Hooks

Use before_* hooks for:

Validation that can block operations
Required field checks
Security policy enforcement
Approval workflows

Use after_* hooks for:

Notifications
Audit logging
External system sync
Non-blocking enrichment

Metadata Storage

Use PluginMetadata for tracking:

return &pb.ObjectHookResponse{
    AllowOperation: true,
    PluginMetadata: map[string]string{
        "validated_by":  "security-plugin",
        "validated_at":  time.Now().Format(time.RFC3339),
        "security_scan": "passed",
        "risk_score":    "low",
    },
}, nil

Multi-Capability Patterns

Shared State

Share data structures between capabilities:

Expandable

type RateLimiterPlugin struct {
    plugin_sdk.BasePlugin
    limits sync.Map // app_id -> limit config
}

// PostAuth uses limits
func (p *RateLimiterPlugin) HandlePostAuth(ctx plugin_sdk.Context, req *pb.EnrichedRequest) (*pb.PluginResponse, error) {
    limit, _ := p.limits.Load(ctx.AppID)
    // Check limit
}

// UI reads limits
func (p *RateLimiterPlugin) HandleCall(method string, payload []byte) ([]byte, error) {
    if method == "get_limits" {
        var limits []LimitInfo
        p.limits.Range(func(key, value interface{}) bool {
            limits = append(limits, LimitInfo{
                AppID: key.(uint32),
                Limit: value.(int),
            })
            return true
        })
        return json.Marshal(limits)
    }
}

State Persistence

Use KV storage for durable state:

Expandable

func (p *MyPlugin) Initialize(ctx plugin_sdk.Context, config map[string]string) error {
    // Load state from KV on startup
    data, err := ctx.Services.KV().Read(ctx, "plugin_state")
    if err == nil {
        json.Unmarshal(data, &p.state)
    }
    return nil
}

func (p *MyPlugin) Shutdown(ctx plugin_sdk.Context) error {
    // Save state to KV on shutdown
    data, _ := json.Marshal(p.state)
    ctx.Services.KV().Write(ctx, "plugin_state", data)
    return nil
}

Deployment

Version Management

Use semantic versioning:

plugin_sdk.NewBasePlugin(
    "my-plugin",
    "1.2.3", // MAJOR.MINOR.PATCH
    "Description",
)

MAJOR: Breaking changes (incompatible API changes)
MINOR: New features (backward-compatible)
PATCH: Bug fixes (backward-compatible)

Resource Cleanup

Always clean up in Shutdown():

Expandable

func (p *MyPlugin) Shutdown(ctx plugin_sdk.Context) error {
    // Close database connections
    if p.db != nil {
        p.db.Close()
    }

    // Close HTTP clients
    if p.httpClient != nil {
        p.httpClient.CloseIdleConnections()
    }

    // Flush any pending operations
    p.flushPendingData(ctx)

    ctx.Services.Logger().Info("Plugin shutdown complete")
    return nil
}

Health Checks

Implement health check methods:

Expandable

func (p *MyPlugin) HandleCall(method string, payload []byte) ([]byte, error) {
    if method == "health" {
        status := map[string]interface{}{
            "status": "healthy",
            "uptime": time.Since(p.startTime).Seconds(),
        }

        // Check external dependencies
        if err := p.checkDependencies(); err != nil {
            status["status"] = "unhealthy"
            status["error"] = err.Error()
        }

        return json.Marshal(status)
    }
    // ... other methods
}

Common Pitfalls

AUTH Plugins: App Linking Requirement

The most common AUTH plugin pitfall is not linking credentials to App objects. A valid credential alone is insufficient—the system needs the App context for access control. Why This Matters: Apps provide the access control context that governs what authenticated requests can do:

Policy enforcement (rate limits, quotas)
Tool and datasource permissions
LLM access restrictions
Budget controls

Required Interface Methods: AUTH plugins must implement all three methods of the AuthHandler interface:

type AuthHandler interface {
    HandleAuth(ctx Context, req *pb.AuthRequest) (*pb.AuthResponse, error)
    GetAppByCredential(ctx Context, credential string) (*pb.App, error)      // Often forgotten!
    GetUserByCredential(ctx Context, credential string) (*pb.User, error)
}

Common Mistakes:

Expandable

// BAD: Only validates token, doesn't return App ID
func (p *MyPlugin) HandleAuth(ctx plugin_sdk.Context, req *pb.AuthRequest) (*pb.AuthResponse, error) {
    if isValidToken(req.Credential) {
        return &pb.AuthResponse{
            Authenticated: true,
            // Missing AppId and UserId! Request will fail.
        }, nil
    }
    return &pb.AuthResponse{Authenticated: false}, nil
}

// GOOD: Returns both App ID and User ID
func (p *MyPlugin) HandleAuth(ctx plugin_sdk.Context, req *pb.AuthRequest) (*pb.AuthResponse, error) {
    tokenConfig, valid := p.lookupToken(req.Credential)
    if !valid {
        return &pb.AuthResponse{Authenticated: false, ErrorMessage: "Invalid token"}, nil
    }
    return &pb.AuthResponse{
        Authenticated: true,
        AppId:         tokenConfig.AppID,   // Links to access control
        UserId:        tokenConfig.UserID,  // Links to identity
    }, nil
}

Validation Checklist:

HandleAuth returns a valid AppId that exists in the database
HandleAuth returns a valid UserId that exists in the database
GetAppByCredential fetches the complete App object via Service API
GetUserByCredential fetches the complete User object via Service API
The App has the required permissions for the tools/LLMs the user needs

See Plugin SDK Reference - AuthHandler for complete documentation.

1. Modifying Shared Data Without Locking

Expandable

// Bad: Race condition
type MyPlugin struct {
    counter int // Not thread-safe!
}

// Good: Use sync primitives
type MyPlugin struct {
    mu      sync.Mutex
    counter int
}

func (p *MyPlugin) increment() {
    p.mu.Lock()
    defer p.mu.Unlock()
    p.counter++
}

2. Blocking in Defer

// Bad: Blocking operations in defer
defer func() {
    p.sendAnalytics(data) // Blocks shutdown!
}()

// Good: Use timeouts
defer func() {
    ctx, cancel := context.WithTimeout(context.Background(), 5*time.Second)
    defer cancel()
    p.sendAnalyticsWithContext(ctx, data)
}()

3. Not Handling Context Cancellation

// Good: Respect context cancellation
func (p *MyPlugin) longRunningOperation(ctx context.Context) error {
    for {
        select {
        case <-ctx.Done():
            return ctx.Err()
        default:
            // Do work
        }
    }
}

4. Excessive Logging

// Bad: Log spam
for _, item := range items {
    ctx.Services.Logger().Debug("Processing item", "item", item)
}

// Good: Batch logging
ctx.Services.Logger().Info("Processing items",
    "count", len(items),
    "batch_id", batchID,
)

General

AI Studio

MCPs

​Availability

​Architecture

​Single Responsibility

​BasePlugin Usage

​Configuration Management

​Error Handling

​Fail Fast, Fail Clearly

​Log Errors with Context

​Graceful Degradation

​Performance

​Minimize Blocking Operations

​Use Connection Pooling

​Cache Frequently Accessed Data

​Batch Operations

​Security

​Input Validation

​Least Privilege Permissions

​Secrets Management

​Sanitize Logs

​Observability

​Structured Logging

​Request Tracing

​Metrics Collection

​Testing

​Unit Tests

​Integration Tests

​Object Hooks Best Practices

​Hook Priority

​Before vs After Hooks

​Metadata Storage

​Multi-Capability Patterns

​Shared State

​State Persistence

​Deployment

​Version Management

​Resource Cleanup

​Health Checks

​Common Pitfalls

​AUTH Plugins: App Linking Requirement

​1. Modifying Shared Data Without Locking

​2. Blocking in Defer

​3. Not Handling Context Cancellation

​4. Excessive Logging

Availability

Architecture

Single Responsibility

BasePlugin Usage

Configuration Management

Error Handling

Fail Fast, Fail Clearly

Log Errors with Context

Graceful Degradation

Performance

Minimize Blocking Operations

Use Connection Pooling

Cache Frequently Accessed Data

Batch Operations

Security

Input Validation

Least Privilege Permissions

Secrets Management

Sanitize Logs

Observability

Structured Logging

Request Tracing

Metrics Collection

Testing

Unit Tests

Integration Tests

Object Hooks Best Practices

Hook Priority

Before vs After Hooks

Metadata Storage

Multi-Capability Patterns

Shared State

State Persistence

Deployment

Version Management

Resource Cleanup

Health Checks

Common Pitfalls

AUTH Plugins: App Linking Requirement

1. Modifying Shared Data Without Locking

2. Blocking in Defer

3. Not Handling Context Cancellation

4. Excessive Logging