MCP Orchestrator

A TypeScript library for orchestrating Model Context Protocol (MCP) servers with AI agents.

Compose multiple MCP servers into powerful workflows using traditional tool calling, LLM sampling, or the innovative code mode that lets LLMs write and execute TypeScript code for complex operations.

Why MCP Orchestrator?

The Code Mode Advantage

While traditional MCP usage exposes tools directly to LLMs through "tool calling" tokens, we offer a better approach for complex workflows:

Traditional Tool Calling:

• ❌ Each tool call requires a full LLM round-trip
• ❌ LLMs only trained on synthetic tool-calling examples
• ❌ Intermediate results waste tokens
• ❌ Complex multi-step operations are inefficient

Code Mode:

• ✅ LLMs write TypeScript code using your tools as an API
• ✅ Chain multiple operations without LLM round-trips
• ✅ 60-75% token reduction for multi-step tasks
• ✅ Generated code is inspectable and reusable

Flexible Orchestration

Choose the right approach for each task:

• Code Mode: Complex workflows, data transformations, multi-step operations
• Tool Calling: Simple, single-step operations
• LLM Sampling: Dual-level sampling (orchestrator and sub-server)
• Composition Patterns: Sequential, parallel, retry, conditional execution

🆕 New in v0.5.0: A2A Integration

Native Agent-to-Agent Support: Expose your orchestrated expert as a compliant A2A agent server.

• Allows other AI agents to discover and collaborate with your expert
• Expose "Code Mode" as a trainable skill
• Bridge any MCP tool to the A2A protocol automatically
• Full support for task lifecycle, streaming, and cancellation

Quick Start

import { MCPOrchestrator } from 'mcp-orchestrator';
import { OpenAIProvider } from 'mcp-orchestrator/llm';

const orchestrator = new MCPOrchestrator({
  servers: {
    filesystem: {
      command: 'npx',
      args: ['-y', '@modelcontextprotocol/server-filesystem', './']
    },
    github: {
      command: 'npx',
      args: ['-y', '@modelcontextprotocol/server-github']
    }
  },
  llm: new OpenAIProvider({ apiKey: process.env.OPENAI_API_KEY })
});

await orchestrator.connect();

// Code Mode: Let LLM write and execute code
const result = await orchestrator.generateAndExecute(
  'Find all TypeScript files, count lines in each, and return the top 5 largest'
);

console.log(result.code);   // See generated code
console.log(result.result); // Get results

Generated code example:

const files = await tools.list_directory({ path: './' });
const tsFiles = files.content.filter(f => f.name.endsWith('.ts'));

const fileSizes = [];
for (const file of tsFiles) {
  const content = await tools.read_file({ path: file.path });
  fileSizes.push({ name: file.name, lines: content.split('\n').length });
}

return fileSizes.sort((a, b) => b.lines - a.lines).slice(0, 5);

Features

Orchestration Capabilities

• 🌐 Multi-Server Management: Connect and coordinate multiple MCP servers (Stdio, HTTP)
• 🎯 Flexible Execution: Choose between code mode, tool calling, or LLM sampling per task
• 🔄 Composition Patterns: Sequential, parallel, retry, and conditional workflows
• 🛠️ Type-Safe Tools: Runtime validation with optional static type generation

Code Mode

• 🚀 LLM Code Generation: AI writes TypeScript code using your tools as an API
• 🔒 Secure Sandbox: VM-based isolation with timeout enforcement
• 🤖 Smart Retry: Automatic error recovery with LLM self-correction
• 📊 Token Efficiency: 60-75% reduction for multi-step operations

Advanced Features

• 📝 Structured Outputs: OpenAI and Anthropic support with Zod validation
• 🔄 Dual-Level Sampling: Orchestrator and sub-server LLM sampling via MCP
• 🤝 A2A Bridge: Native Agent-to-Agent protocol support for multi-agent collaboration
• 🔧 Zero-Opinion: Use with any workflow engine or plain async/await
• 🔐 Security: Approval workflows, audit logging, rate limiting

Installation

npm install mcp-orchestrator-ts

Direct Code Execution

You can also write and execute your own code directly:

const code = `
  const files = await tools.list_directory({ path: './' });
  const tsFiles = files.content.filter(f => f.name.endsWith('.ts'));
  
  const fileSizes = [];
  for (const file of tsFiles) {
    const content = await tools.read_file({ path: file.path });
    fileSizes.push({ 
      name: file.name, 
      lines: content.split('\\n').length 
    });
  }
  
  return fileSizes.sort((a, b) => b.lines - a.lines).slice(0, 5);
`;

const result = await orchestrator.executeCode(code, {
  timeout: 30000
});
});

Snippet System (Code Reuse)

Promote useful "Code Mode" scripts to first-class MCP tools automatically.

1. Enable Snippet Mode

Set environment variables:

ENABLE_SNIPPET_MODE=true
SNIPPET_STORAGE_PATH=./snippets  # Optional, defaults to ./snippets

2. Create a Snippet

Instruct the LLM to include metadata comments in the generated code:

// @name: my-tool-name
// @description: A description of what the tool does
// @input: {"type": "object", "properties": {"arg1": {"type": "string"}}}

// Your code using 'args'
console.log(args.arg1);

3. Use the Snippet

The snippet is automatically saved and registered as a tool. You can now use it like any other MCP tool:

await orchestrator.callTool('my-tool-name', { arg1: 'value' });

4. Auto-Saving

If you use generateAndExecute with saveToSnippets: true, the orchestrator will automatically save any code with valid metadata that executes successfully.

await orchestrator.generateAndExecute('Create a tool named "echo" that prints the input', {
  saveToSnippets: true
});

A2A Integration (Agent-to-Agent)

Expose your orchestrated expert as a native A2A agent, allowing other agents to collaborate with it.

1. Create the Bridge

import { McpA2aBridge } from 'mcp-orchestrator/a2a';

// Create a bridge for your orchestrator
const bridge = new McpA2aBridge(orchestrator, {
    name: "My Expert Agent",
    description: "An agent that helps with X, Y, and Z"
});

2. Add Skills

You can expose specific capabilities as A2A "Skills":

// Expose the "Code Mode" engine (Natural Language -> Code -> Execution)
bridge.addCodeModeSkill({
    name: "Code Expert",
    systemPrompt: "You are a helpful coding assistant."
});

// Expose a specific tool directly
bridge.addToolSkill("calculate_sum");

3. Mount to Express

const app = express();
app.use(express.json());

// Mount the A2A protocol endpoints
app.use("/a2a", bridge.createExpressRouter());

Your agent is now discoverable via /.well-known/agent-card.json and supports the full A2A lifecycle (tasks, status updates, cancellation).

Traditional Tool Calling

For simple, single-step operations, you can still use traditional tool calling:

// Direct tool call
const files = await orchestrator.callTool('list_directory', { path: './' });

// With structured LLM output
const analysis = await orchestrator.llm.generateStructured({
  schema: z.object({ 
    summary: z.string(),
    fileCount: z.number()
  }),
  prompt: `Analyze these files: ${JSON.stringify(files)}`
});

When to Use Each Approach

Use Code Mode	Use Traditional Tool Calling
Multi-step workflows	Single tool calls
Complex data transformations	Simple operations
Many tools available	Few tools
Performance matters (60-75% token savings)	Real-time interaction

📖 See docs/code-mode.md for comprehensive code mode documentation.

Dual-Level LLM Sampling

The library supports two levels of LLM sampling through the Model Context Protocol (MCP):

1. Orchestrator-Level Sampling

Direct LLM sampling via MCP sampling/createMessage when supported by MCP clients.

// Enable sampling with default options
const orchestrator = new MCPOrchestrator({
  servers: { /* your servers */ },
  llm: new OpenAIProvider({ apiKey: process.env.OPENAI_API_KEY }),
  samplingOptions: {
    maxTokens: 1000,
    temperature: 0.7,
    requireApproval: true, // Require user approval for sampling requests
    timeoutMs: 30000
  }
});

// Check sampling capabilities
const capabilities = orchestrator.getSamplingCapabilities();
console.log('Client supports sampling:', capabilities);

// Perform orchestrator-level sampling
const result = await orchestrator.sample(
  [
    { role: 'user', content: 'Analyze the benefits of MCP orchestration' }
  ],
  {
    systemPrompt: 'You are an expert in AI and distributed systems.',
    maxTokens: 500,
    origin: 'orchestrator'
  }
);

console.log('Sample result:', result);

2. Sub-Server Sampling via Proxy

Lightweight sampling proxy that forwards sub-server requests to the orchestrator's LLM provider.

// Create sampling proxy for a sub-server
const expertProxy = orchestrator.createSamplingProxy('data-analyzer', {
  origin: 'data-expert',
  maxTokens: 300,
  requireApproval: false // Sub-server requests typically don't need approval
});

// Sub-server makes sampling request
const analysisResult = await expertProxy.createMessage({
  messages: [
    { role: 'user', content: 'Analyze this dataset for trends' },
    { role: 'assistant', content: 'I\'ll analyze the data for you.' }
  ],
  systemPrompt: 'You are a data analysis expert.',
  maxTokens: 400
});

// Structured sampling for type-safe results
const structuredResult = await expertProxy.createMessageStructured(
  {
    messages: [{ role: 'user', content: 'Provide analysis in JSON format' }],
    systemPrompt: 'Return structured analysis results.',
  },
  undefined,
  z.object({
    trends: z.array(z.string()),
    confidence: z.number(),
    recommendations: z.array(z.string())
  })
);

Security & Trust Features

Built-in security and trust management for LLM sampling:

import { SamplingSecurityManager, SecurityPolicy } from 'mcp-orchestrator/sampling';

// Configure security manager
const securityManager = new SamplingSecurityManager({
  maxQueueSize: 50,
  defaultRateLimit: {
    requestsPerMinute: 30,
    requestsPerHour: 500,
    requestsPerDay: 5000,
  }
});

// Add custom security policy
const costLimitPolicy: SecurityPolicy = {
  name: 'cost_limit_policy',
  description: 'Reject expensive requests',
  evaluate: async (request, options, context) => {
    const estimatedCost = estimateCost(request);
    if (estimatedCost > 1.00) {
      return { approved: false, reason: 'Cost exceeds limit' };
    }
    return { approved: true };
  }
};

securityManager.addPolicy(costLimitPolicy);

// Handle approval workflow (for UI integration)
securityManager.on('approval_requested', (approval) => {
  console.log('User approval needed:', approval.id);
  // Show approval dialog in your UI
});

// Check rate limits
const rateLimitStatus = securityManager.getRateLimitStatus('my-app');
console.log('Rate limit status:', rateLimitStatus);

// Access audit log
const auditLog = securityManager.getAuditLog({
  eventType: 'approved',
  startTime: Date.now() - 3600000 // Last hour
});

Tool-Enabled Sampling

Support for LLM-to-tool reasoning loops when both client and server support it:

// Define available tools for sampling
const tools = [
  {
    name: 'search_database',
    description: 'Search the database for information',
    inputSchema: {
      type: 'object',
      properties: {
        query: { type: 'string' },
        limit: { type: 'number' }
      },
      required: ['query']
    }
  },
  {
    name: 'calculate_metrics',
    description: 'Calculate performance metrics',
    inputSchema: {
      type: 'object',
      properties: {
        data: { type: 'array' },
        operation: { type: 'string' }
      },
      required: ['data']
    }
  }
];

// Use tool-enabled sampling
const result = await orchestrator.sample(
  [
    { 
      role: 'user', 
      content: 'Find relevant data and calculate average performance' 
    }
  ],
  {
    tools,
    toolChoice: 'auto', // Let LLM choose which tools to use
    origin: 'analyst-expert'
  }
);

// The LLM will decide when to use tools and the system will handle the tool loop

Type Generation

Generate TypeScript types for your tools to get full IDE autocomplete and type safety.

1. Create a Configuration File

Create a mcp-config.json file that defines your servers. This file supports both stdio (command-line tools) and sse (remote HTTP) servers.

{
  "servers": {
    "filesystem": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "./"]
    },
    "github": {
      "url": "https://api.github.com/mcp",
      "auth": {
        "token": "YOUR_GITHUB_TOKEN"
      }
    },
    "aws-tools": {
      "command": "node",
      "args": ["./dist/index.js"],
      "env": {
        "AWS_REGION": "us-east-1"
      }
    }
  }
}

2. Run the Generator

Run the CLI tool to connect to the servers, discover tools, and generate a TypeScript file with interfaces.

npx mcp-orchestrator --config mcp-config.json --output src/types/mcp-tools.ts

3. Use Generated Types

Import the generated types to ensure type safety when calling tools.

import { MCPOrchestrator } from 'mcp-orchestrator';
import './types/mcp-tools'; // Load type augmentations

const orchestrator = new MCPOrchestrator({ /* ... */ });

// TypeScript will now validate arguments and return types!
const result = await orchestrator.callTool('aws_ec2_run_instances', {
  instanceType: 't3.medium', // Type-checked
  region: 'us-east-1'
});

Expert Composition Patterns

The library provides helper functions to compose tools into complex workflows.

Sequential Execution

Pass state through a series of steps.

import { sequence } from 'mcp-orchestrator/patterns';

const result = await sequence([
  async (ctx) => {
    const data = await orchestrator.callTool('fetch_data', { id: ctx.id });
    return { ...ctx, data };
  },
  async (ctx) => {
    const analysis = await orchestrator.llm.generateStructured({
      schema: AnalysisSchema,
      prompt: `Analyze: ${JSON.stringify(ctx.data)}`
    });
    return { ...ctx, analysis };
  }
], { id: '123' });

Parallel Execution

Run multiple independent tasks concurrently.

import { parallel } from 'mcp-orchestrator/patterns';

const [users, posts] = await parallel([
  () => orchestrator.callTool('get_users', {}),
  () => orchestrator.callTool('get_posts', {})
]);

Retry Logic

Automatically retry transient failures with backoff.

import { retry } from 'mcp-orchestrator/patterns';

const result = await retry(
  () => orchestrator.callTool('flaky_api', {}),
  { 
    maxAttempts: 3, 
    backoff: 'exponential',
    initialDelay: 1000 
  }
);

Best Practices

Sampling Considerations

• Cost Management: Use security policies to limit expensive requests
• User Approval: Enable approval workflows for sensitive operations
• Rate Limiting: Implement appropriate rate limits to prevent abuse
• Fallback Strategy: Always provide fallback when MCP sampling isn't available

Security Recommendations

• Origin Tracking: Use distinct origins for different components to track usage
• Audit Logging: Regularly review audit logs for unusual patterns
• Policy Enforcement: Implement content filtering and cost control policies
• Timeout Handling: Set appropriate timeouts to prevent hanging requests

Tool-Enabled Sampling

• Tool Design: Keep tool schemas simple and well-documented
• Error Handling: Implement robust error handling for tool execution
• Loop Prevention: Set reasonable limits to prevent infinite tool loops
• User Experience: Provide clear feedback when tool-enabled sampling is used

License

MIT

Support my work

You can support my work by signing up for;

• Sinay.ai