MCP server optimization and transport layer enhancement for claude-flow v3. Implements connection pooling, load balancing, tool registry optimization, and performance monitoring for sub-100ms response times.
npx skills add https://github.com/spencermarx/open-code-review --skill V3 MCP Optimization
Optimizes claude-flow v3 MCP (Model Context Protocol) server implementation with advanced transport layer optimizations, connection pooling, load balancing, and comprehensive performance monitoring to achieve sub-100ms response times.
# Initialize MCP optimization analysis
Task("MCP architecture", "Analyze current MCP server performance and bottlenecks", "mcp-specialist")
# Optimization implementation (parallel)
Task("Connection pooling", "Implement MCP connection pooling and reuse", "mcp-specialist")
Task("Load balancing", "Add dynamic load balancing for MCP tools", "mcp-specialist")
Task("Transport optimization", "Optimize transport layer performance", "mcp-specialist")
Current MCP Issues:
├── Cold Start Latency: ~1.8s MCP server init
├── Connection Overhead: New connection per request
├── Tool Registry: Linear search O(n) for 213+ tools
├── Transport Layer: No connection reuse
└── Memory Usage: No cleanup of idle connections
Target Performance:
├── Startup Time: <400ms (4.5x improvement)
├── Tool Lookup: <5ms (O(1) hash table)
├── Connection Reuse: 90%+ connection pool hits
├── Response Time: <100ms p95
└── Memory Efficiency: 50% reduction
// src/core/mcp/mcp-server.ts
import { Server } from '@modelcontextprotocol/sdk/server/index.js';
import { StdioServerTransport } from '@modelcontextprotocol/sdk/server/stdio.js';
interface OptimizedMCPConfig {
// Connection pooling
maxConnections: number;
idleTimeoutMs: number;
connectionReuseEnabled: boolean;
// Tool registry
toolCacheEnabled: boolean;
toolIndexType: 'hash' | 'trie';
// Performance
requestTimeoutMs: number;
batchingEnabled: boolean;
compressionEnabled: boolean;
// Monitoring
metricsEnabled: boolean;
healthCheckIntervalMs: number;
}
export class OptimizedMCPServer {
private server: Server;
private connectionPool: ConnectionPool;
private toolRegistry: FastToolRegistry;
private loadBalancer: MCPLoadBalancer;
private metrics: MCPMetrics;
constructor(config: OptimizedMCPConfig) {
this.server = new Server({
name: 'claude-flow-v3',
version: '3.0.0'
}, {
capabilities: {
tools: { listChanged: true },
resources: { subscribe: true, listChanged: true },
prompts: { listChanged: true }
}
});
this.connectionPool = new ConnectionPool(config);
this.toolRegistry = new FastToolRegistry(config.toolIndexType);
this.loadBalancer = new MCPLoadBalancer();
this.metrics = new MCPMetrics(config.metricsEnabled);
}
async start(): Promise<void> {
// Pre-warm connection pool
await this.connectionPool.preWarm();
// Pre-build tool index
await this.toolRegistry.buildIndex();
// Setup request handlers with optimizations
this.setupOptimizedHandlers();
// Start health monitoring
this.startHealthMonitoring();
// Start server
const transport = new StdioServerTransport();
await this.server.connect(transport);
this.metrics.recordStartup();
}
}
// src/core/mcp/connection-pool.ts
interface PooledConnection {
id: string;
connection: MCPConnection;
lastUsed: number;
usageCount: number;
isHealthy: boolean;
}
export class ConnectionPool {
private pool: Map<string, PooledConnection> = new Map();
private readonly config: ConnectionPoolConfig;
private healthChecker: HealthChecker;
constructor(config: ConnectionPoolConfig) {
this.config = {
maxConnections: 50,
minConnections: 5,
idleTimeoutMs: 300000, // 5 minutes
maxUsageCount: 1000,
healthCheckIntervalMs: 30000,
...config
};
this.healthChecker = new HealthChecker(this.config.healthCheckIntervalMs);
}
async getConnection(endpoint: string): Promise<MCPConnection> {
const start = performance.now();
// Try to get from pool first
const pooled = this.findAvailableConnection(endpoint);
if (pooled) {
pooled.lastUsed = Date.now();
pooled.usageCount++;
this.recordMetric('pool_hit', performance.now() - start);
return pooled.connection;
}
// Check pool capacity
if (this.pool.size >= this.config.maxConnections) {
await this.evictLeastUsedConnection();
}
// Create new connection
const connection = await this.createConnection(endpoint);
const pooledConn: PooledConnection = {
id: this.generateConnectionId(),
connection,
lastUsed: Date.now(),
usageCount: 1,
isHealthy: true
};
this.pool.set(pooledConn.id, pooledConn);
this.recordMetric('pool_miss', performance.now() - start);
return connection;
}
async releaseConnection(connection: MCPConnection): Promise<void> {
// Mark connection as available for reuse
const pooled = this.findConnectionById(connection.id);
if (pooled) {
// Check if connection should be retired
if (pooled.usageCount >= this.config.maxUsageCount) {
await this.removeConnection(pooled.id);
}
}
}
async preWarm(): Promise<void> {
const connections: Promise<MCPConnection>[] = [];
for (let i = 0; i < this.config.minConnections; i++) {
connections.push(this.createConnection('default'));
}
await Promise.all(connections);
}
private async evictLeastUsedConnection(): Promise<void> {
let oldestConn: PooledConnection | null = null;
let oldestTime = Date.now();
for (const conn of this.pool.values()) {
if (conn.lastUsed < oldestTime) {
oldestTime = conn.lastUsed;
oldestConn = conn;
}
}
if (oldestConn) {
await this.removeConnection(oldestConn.id);
}
}
private findAvailableConnection(endpoint: string): PooledConnection | null {
for (const conn of this.pool.values()) {
if (conn.isHealthy &&
conn.connection.endpoint === endpoint &&
Date.now() - conn.lastUsed < this.config.idleTimeoutMs) {
return conn;
}
}
return null;
}
}
// src/core/mcp/fast-tool-registry.ts
interface ToolIndexEntry {
name: string;
handler: ToolHandler;
metadata: ToolMetadata;
usageCount: number;
avgLatencyMs: number;
}
export class FastToolRegistry {
private toolIndex: Map<string, ToolIndexEntry> = new Map();
private categoryIndex: Map<string, string[]> = new Map();
private fuzzyMatcher: FuzzyMatcher;
private cache: LRUCache<string, ToolIndexEntry>;
constructor(indexType: 'hash' | 'trie' = 'hash') {
this.fuzzyMatcher = new FuzzyMatcher();
this.cache = new LRUCache<string, ToolIndexEntry>(1000); // Cache 1000 most used tools
}
async buildIndex(): Promise<void> {
const start = performance.now();
// Load all available tools
const tools = await this.loadAllTools();
// Build hash index for O(1) lookup
for (const tool of tools) {
const entry: ToolIndexEntry = {
name: tool.name,
handler: tool.handler,
metadata: tool.metadata,
usageCount: 0,
avgLatencyMs: 0
};
this.toolIndex.set(tool.name, entry);
// Build category index
const category = tool.metadata.category || 'general';
if (!this.categoryIndex.has(category)) {
this.categoryIndex.set(category, []);
}
this.categoryIndex.get(category)!.push(tool.name);
}
// Build fuzzy search index
await this.fuzzyMatcher.buildIndex(tools.map(t => t.name));
console.log(`Tool index built in ${(performance.now() - start).toFixed(2)}ms for ${tools.length} tools`);
}
findTool(name: string): ToolIndexEntry | null {
// Try cache first
const cached = this.cache.get(name);
if (cached) return cached;
// Try exact match
const exact = this.toolIndex.get(name);
if (exact) {
this.cache.set(name, exact);
return exact;
}
// Try fuzzy match
const fuzzyMatches = this.fuzzyMatcher.search(name, 1);
if (fuzzyMatches.length > 0) {
const match = this.toolIndex.get(fuzzyMatches[0]);
if (match) {
this.cache.set(name, match);
return match;
}
}
return null;
}
findToolsByCategory(category: string): ToolIndexEntry[] {
const toolNames = this.categoryIndex.get(category) || [];
return toolNames
.map(name => this.toolIndex.get(name))
.filter(entry => entry !== undefined) as ToolIndexEntry[];
}
getMostUsedTools(limit: number = 10): ToolIndexEntry[] {
return Array.from(this.toolIndex.values())
.sort((a, b) => b.usageCount - a.usageCount)
.slice(0, limit);
}
recordToolUsage(toolName: string, latencyMs: number): void {
const entry = this.toolIndex.get(toolName);
if (entry) {
entry.usageCount++;
// Moving average for latency
entry.avgLatencyMs = (entry.avgLatencyMs + latencyMs) / 2;
}
}
}
// src/core/mcp/load-balancer.ts
interface ServerInstance {
id: string;
endpoint: string;
load: number;
responseTime: number;
isHealthy: boolean;
maxConnections: number;
currentConnections: number;
}
export class MCPLoadBalancer {
private servers: Map<string, ServerInstance> = new Map();
private routingStrategy: RoutingStrategy = 'least-connections';
addServer(server: ServerInstance): void {
this.servers.set(server.id, server);
}
selectServer(toolCategory?: string): ServerInstance | null {
const healthyServers = Array.from(this.servers.values())
.filter(server => server.isHealthy);
if (healthyServers.length === 0) return null;
switch (this.routingStrategy) {
case 'round-robin':
return this.roundRobinSelection(healthyServers);
case 'least-connections':
return this.leastConnectionsSelection(healthyServers);
case 'response-time':
return this.responseTimeSelection(healthyServers);
case 'weighted':
return this.weightedSelection(healthyServers, toolCategory);
default:
return healthyServers[0];
}
}
private leastConnectionsSelection(servers: ServerInstance[]): ServerInstance {
return servers.reduce((least, current) =>
current.currentConnections < least.currentConnections ? current : least
);
}
private responseTimeSelection(servers: ServerInstance[]): ServerInstance {
return servers.reduce((fastest, current) =>
current.responseTime < fastest.responseTime ? current : fastest
);
}
private weightedSelection(servers: ServerInstance[], category?: string): ServerInstance {
// Prefer servers with lower load and better response time
const scored = servers.map(server => ({
server,
score: this.calculateServerScore(server, category)
}));
scored.sort((a, b) => b.score - a.score);
return scored[0].server;
}
private calculateServerScore(server: ServerInstance, category?: string): number {
const loadFactor = 1 - (server.currentConnections / server.maxConnections);
const responseFactor = 1 / (server.responseTime + 1);
const categoryBonus = this.getCategoryBonus(server, category);
return loadFactor * 0.4 + responseFactor * 0.4 + categoryBonus * 0.2;
}
updateServerMetrics(serverId: string, metrics: Partial<ServerInstance>): void {
const server = this.servers.get(serverId);
if (server) {
Object.assign(server, metrics);
}
}
}
// src/core/mcp/optimized-transport.ts
export class OptimizedTransport {
private compression: boolean = true;
private batching: boolean = true;
private batchBuffer: MCPMessage[] = [];
private batchTimeout: NodeJS.Timeout | null = null;
constructor(private config: TransportConfig) {}
async send(message: MCPMessage): Promise<void> {
if (this.batching && this.canBatch(message)) {
this.addToBatch(message);
return;
}
await this.sendImmediate(message);
}
private async sendImmediate(message: MCPMessage): Promise<void> {
const start = performance.now();
// Compress if enabled
const payload = this.compression
? await this.compress(message)
: message;
// Send through transport
await this.transport.send(payload);
// Record metrics
this.recordLatency(performance.now() - start);
}
private addToBatch(message: MCPMessage): void {
this.batchBuffer.push(message);
// Start batch timeout if not already running
if (!this.batchTimeout) {
this.batchTimeout = setTimeout(
() => this.flushBatch(),
this.config.batchTimeoutMs || 10
);
}
// Flush if batch is full
if (this.batchBuffer.length >= this.config.maxBatchSize) {
this.flushBatch();
}
}
private async flushBatch(): Promise<void> {
if (this.batchBuffer.length === 0) return;
const batch = this.batchBuffer.splice(0);
this.batchTimeout = null;
// Send as single batched message
await this.sendImmediate({
type: 'batch',
messages: batch
});
}
private canBatch(message: MCPMessage): boolean {
// Don't batch urgent messages or responses
return message.type !== 'response' &&
message.priority !== 'high' &&
message.type !== 'error';
}
private async compress(data: any): Promise<Buffer> {
// Use fast compression for smaller messages
return gzipSync(JSON.stringify(data));
}
}
// src/core/mcp/metrics.ts
interface MCPMetrics {
requestCount: number;
errorCount: number;
avgResponseTime: number;
p95ResponseTime: number;
connectionPoolHits: number;
connectionPoolMisses: number;
toolLookupTime: number;
startupTime: number;
}
export class MCPMetricsCollector {
private metrics: MCPMetrics;
private responseTimeBuffer: number[] = [];
private readonly bufferSize = 1000;
constructor() {
this.metrics = this.createInitialMetrics();
}
recordRequest(latencyMs: number): void {
this.metrics.requestCount++;
this.updateResponseTimes(latencyMs);
}
recordError(): void {
this.metrics.errorCount++;
}
recordConnectionPoolHit(): void {
this.metrics.connectionPoolHits++;
}
recordConnectionPoolMiss(): void {
this.metrics.connectionPoolMisses++;
}
recordToolLookup(latencyMs: number): void {
this.metrics.toolLookupTime = this.updateMovingAverage(
this.metrics.toolLookupTime,
latencyMs
);
}
recordStartup(latencyMs: number): void {
this.metrics.startupTime = latencyMs;
}
getMetrics(): MCPMetrics {
return { ...this.metrics };
}
getHealthStatus(): HealthStatus {
const errorRate = this.metrics.errorCount / this.metrics.requestCount;
const poolHitRate = this.metrics.connectionPoolHits /
(this.metrics.connectionPoolHits + this.metrics.connectionPoolMisses);
return {
status: this.determineHealthStatus(errorRate, poolHitRate),
errorRate,
poolHitRate,
avgResponseTime: this.metrics.avgResponseTime,
p95ResponseTime: this.metrics.p95ResponseTime
};
}
private updateResponseTimes(latency: number): void {
this.responseTimeBuffer.push(latency);
if (this.responseTimeBuffer.length > this.bufferSize) {
this.responseTimeBuffer.shift();
}
this.metrics.avgResponseTime = this.calculateAverage(this.responseTimeBuffer);
this.metrics.p95ResponseTime = this.calculatePercentile(this.responseTimeBuffer, 95);
}
private calculatePercentile(arr: number[], percentile: number): number {
const sorted = arr.slice().sort((a, b) => a - b);
const index = Math.ceil((percentile / 100) * sorted.length) - 1;
return sorted[index] || 0;
}
private determineHealthStatus(errorRate: number, poolHitRate: number): 'healthy' | 'warning' | 'critical' {
if (errorRate > 0.1 || poolHitRate < 0.5) return 'critical';
if (errorRate > 0.05 || poolHitRate < 0.7) return 'warning';
return 'healthy';
}
}
// src/core/mcp/tool-precompiler.ts
export class ToolPrecompiler {
async precompileTools(): Promise<CompiledToolRegistry> {
const tools = await this.loadAllTools();
// Create optimized lookup structures
const nameIndex = new Map<string, Tool>();
const categoryIndex = new Map<string, Tool[]>();
const fuzzyIndex = new Map<string, string[]>();
for (const tool of tools) {
// Exact name index
nameIndex.set(tool.name, tool);
// Category index
const category = tool.metadata.category || 'general';
if (!categoryIndex.has(category)) {
categoryIndex.set(category, []);
}
categoryIndex.get(category)!.push(tool);
// Pre-compute fuzzy variations
const variations = this.generateFuzzyVariations(tool.name);
for (const variation of variations) {
if (!fuzzyIndex.has(variation)) {
fuzzyIndex.set(variation, []);
}
fuzzyIndex.get(variation)!.push(tool.name);
}
}
return {
nameIndex,
categoryIndex,
fuzzyIndex,
totalTools: tools.length,
compiledAt: new Date()
};
}
private generateFuzzyVariations(name: string): string[] {
const variations: string[] = [];
// Common typos and abbreviations
variations.push(name.toLowerCase());
variations.push(name.replace(/[-_]/g, ''));
variations.push(name.replace(/[aeiou]/gi, '')); // Consonants only
// Add more fuzzy matching logic as needed
return variations;
}
}
// src/core/mcp/multi-level-cache.ts
export class MultiLevelCache {
private l1Cache: Map<string, any> = new Map(); // In-memory, fastest
private l2Cache: LRUCache<string, any>; // LRU cache, larger capacity
private l3Cache: DiskCache; // Persistent disk cache
constructor(config: CacheConfig) {
this.l2Cache = new LRUCache<string, any>({
max: config.l2MaxEntries || 10000,
ttl: config.l2TTL || 300000 // 5 minutes
});
this.l3Cache = new DiskCache(config.l3Path || './.cache/mcp');
}
async get(key: string): Promise<any | null> {
// Try L1 cache first (fastest)
if (this.l1Cache.has(key)) {
return this.l1Cache.get(key);
}
// Try L2 cache
const l2Value = this.l2Cache.get(key);
if (l2Value) {
// Promote to L1
this.l1Cache.set(key, l2Value);
return l2Value;
}
// Try L3 cache (disk)
const l3Value = await this.l3Cache.get(key);
if (l3Value) {
// Promote to L2 and L1
this.l2Cache.set(key, l3Value);
this.l1Cache.set(key, l3Value);
return l3Value;
}
return null;
}
async set(key: string, value: any, options?: CacheOptions): Promise<void> {
// Set in all levels
this.l1Cache.set(key, value);
this.l2Cache.set(key, value);
if (options?.persistent) {
await this.l3Cache.set(key, value);
}
// Manage L1 cache size
if (this.l1Cache.size > 1000) {
const firstKey = this.l1Cache.keys().next().value;
this.l1Cache.delete(firstKey);
}
}
}
const mcpDashboard = {
metrics: [
'Request latency (p50, p95, p99)',
'Error rate by tool category',
'Connection pool utilization',
'Tool lookup performance',
'Memory usage trends',
'Cache hit rates (L1, L2, L3)'
],
alerts: [
'Response time >200ms for 5 minutes',
'Error rate >5% for 1 minute',
'Pool hit rate <70% for 10 minutes',
'Memory usage >500MB for 5 minutes'
]
};
v3-core-implementation - Core domain integration with MCPv3-performance-optimization - Overall performance optimizationv3-swarm-coordination - MCP integration with swarm coordinationv3-memory-unification - Memory sharing via MCP tools# Full MCP server optimization
Task("MCP optimization implementation",
"Implement all MCP performance optimizations with monitoring",
"mcp-specialist")
# Connection pool optimization
Task("MCP connection pooling",
"Implement advanced connection pooling with health monitoring",
"mcp-specialist")
Skill converted from mcp-deploy-manage-agents.prompt.md
Use this skill when the user wants to launch a new AltClaw, OpenClaw, PicoClaw, or Ottie deployment through Cloud Claw. Covers the same user-facing fields and constraints exposed in the Cloud Claw UI, using the local altllm cloud-claw-* commands. Do NOT use for post-launch lifecycle tasks like start/stop/delete/logs; use cloud-claw-manage-vm.
Build hosted agents using Azure AI Projects SDK with ImageBasedHostedAgentDefinition. Use when creating container-based agents in Azure AI Foundry.
Build MCP (Model Context Protocol) servers on Cloudflare Workers with tools, resources, and prompts.
Chain agent outputs as inputs in sequential or parallel pipelines for data flow orchestration
Audit cloned or reimplemented websites for fidelity gaps, tracking scripts, source-brand and language residue, placeholders, and risky external dependencies. Use before handoff or deployment, or when asked to review a website clone for cleanup and readiness.
> Install and operate Hermes Tweet, a Hermes Agent plugin for X/Twitter research, timeline reading, tweet analysis, and approval-gated tweet actions. Use this skill when installing Hermes Tweet, researching X/Twitter accounts, monitoring launch signals, investigating mentions, auditing giveaways, or preparing guarded tweet actions. Use proactively when a Hermes Agent workflow needs current X/Twitter context. Requires XQUIK_API_KEY for read and action tools.
Handles LLM-as-judge evaluation workflows on Arize including creating/updating evaluators, running evaluations on spans or experiments, managing tasks, trigger-run operations, column mapping, and continuous monitoring. Use when the user mentions create evaluator, LLM judge, hallucination, faithfulness, correctness, relevance, run eval, score spans, score experiment, trigger-run, column mapping, continuous monitoring, or improve evaluator prompt.
Take spencermarx/v3 mcp optimization from the repository into ~/.claude/skills for personal
use, or into .claude/skills inside a project.
The agent identifies a skill by the name field in its header. Two skills with the
same name cannot sit side by side — one of them will be ignored.