MCPC

Build agentic MCP servers by composing existing MCP tools.

MCPC is the SDK for building agentic MCP (Model Context Protocol) Servers. You can use it to:

1. Create Powerful Agentic MCP Tools: Simply describe your vision in text and reference tools from the expanding MCP community. As standard MCP tools, your agents work everywhere and collaborate seamlessly.
2. Fine-Tune Existing Tools: Flexibly modify existing tool descriptions and parameters, or wrap and filter results to precisely adapt them to your specific business scenarios.
3. Build Multi-Agent Systems: By defining each agent as a MCP tool, you can compose and orchestrate them to construct sophisticated, collaborative multi-agent systems.

Key Features

• Portability and agent interoperability: Build once, run everywhere as MCP tools - agents work across all MCP clients and can discover and collaborate with each other through standard MCP interfaces
• Simple composition and fine-tuning: Compose MCP servers as building blocks, select and customize tools, or modify their descriptions and parameters
• Logging and tracing: Built-in MCP logging and OpenTelemetry tracing support
• Skills support: Define domain-specific knowledge following the Agent Skills specification - deploy to production, share via MCP, and declare tool dependencies
• Flexible execution modes: Multiple specialized modes to fit different scenarios - interactive agent (agentic), AI SDK sampling (ai_sampling), AI ACP mode (ai_acp), secure code execution (code_execution), and sandbox + sampling (code_execution_sampling) - each with dedicated implementations

Quick Start

Three Ways to Get Started

1. Use the Website (Fastest)

Visit mcpc.tech to browse servers from the official MCP registry, discover tools, and generate ready-to-use agents.

2. Use the Agent (Interactive)

Let AI help you discover servers and build agents:

Add to your MCP client:

{
  "mcpServers": {
    "mcpc-builder-agent": {
      "command": "npx",
      "args": ["-y", "@mcpc-tech/builder", "mcpc-builder-agent"]
    }
  }
}

3. Write Code (Full Control)

Use the SDK directly for complete customization. See examples below.

Installation

# npm (from npm registry)
npm install @mcpc-tech/core
# npm (from jsr)
npx jsr add @mcpc/core

# deno
deno add jsr:@mcpc/core

# pnpm (from npm registry)
pnpm add @mcpc-tech/core
# pnpm (from jsr)
pnpm add jsr:@mcpc/core

Or run directly with the CLI (no installation required):

# Run with remote configuration
npx -y @mcpc-tech/cli --config-url \
  "https://raw.githubusercontent.com/mcpc-tech/mcpc/main/packages/cli/examples/configs/codex-fork.json"

Examples: Create a Simple Codex/Claude Code Fork

import { mcpc } from "@mcpc/core";

const server = await mcpc(
  [{ name: "coding-agent", version: "0.1.0" }, { capabilities: { tools: {} } }],
  [{
    name: "coding-agent",
    description: `
      You are a coding assistant with advanced capabilities.

      Your capabilities include:
      - Reading and writing files
      - Executing terminal commands to build, test, and run projects
      - Interacting with GitHub to create pull requests and manage issues

      Available tools:
      <tool name="desktop-commander.execute_command" />
      <tool name="desktop-commander.read_file" />
      <tool name="desktop-commander.write_file" />
      <tool name="github.create_pull_request" />
    `,
    deps: {
      mcpServers: {
        "desktop-commander": {
          command: "npx",
          args: ["-y", "@wonderwhy-er/desktop-commander@latest"],
          transportType: "stdio",
        },
        github: {
          transportType: "streamable-http",
          url: "https://api.githubcopilot.com/mcp/",
        },
      },
    },
  }],
);

Complete Example: See the full Codex fork tutorial.

Install the MCPC Core Skill

Install the mcpc-core skill into your project using skills.sh:

npx skills add mcpc-tech/mcpc

This installs the skill into .agents/skills/mcpc-core/, giving your agent on-demand access to the full @mcpc/core API reference, usage patterns, plugin guide, and gotchas.

Examples: Load Agent Skills

For complex agents where inline description becomes unwieldy, use Agent Skills to organize domain knowledge in separate files that are loaded on-demand.

import { createBashPlugin, createSkillsPlugin } from "@mcpc/core/plugins";

const server = await mcpc(
  [{ name: "my-agent", version: "1.0.0" }, { capabilities: { tools: {} } }],
  [{
    name: "my-agent",
    description: 'An agent with domain knowledge\n\n<tool name="bash"/>',
    plugins: [
      createSkillsPlugin({ paths: ["./skills"] }),
      createBashPlugin(),
    ],
  }],
);

Skills load domain knowledge on-demand, while bash plugin enables script execution. For scripts in scripts/ directory, skills returns the path - use bash tool to execute.

Complete Examples: See 14-skills-plugin.ts and 25-skills-with-bash.ts.

Examples: Progressive Manual Disclosure

For agents with detailed instructions, use the manual field to reduce initial prompt length - the full manual is fetched on-demand via man { manual: true }:

const server = await mcpc(
  [{ name: "code-reviewer", version: "1.0.0" }, {
    capabilities: { tools: {} },
  }],
  [{
    name: "code-reviewer",
    description: "AI code reviewer for quality and security analysis.",
    manual: `Detailed review guidelines...
<tool name="desktop-commander.read_file"/>
<tool name="desktop-commander.write_file"/>

## Review Categories
1. Code Quality - readability, naming, complexity
2. Security - SQL injection, XSS, credentials
...`,
    deps: {
      mcpServers: {
        "desktop-commander": {
          command: "npx",
          args: ["-y", "@wonderwhy-er/desktop-commander@0.1.20"],
          transportType: "stdio",
        },
      },
    },
  }],
);

Complete Example: See 21-progressive-manual.ts.

How It Works

Three simple steps:

1. Define dependencies - List the MCP servers you want to use
2. Write agent description - Describe what your agent does and reference tools
3. Create server - Use mcpc() to build and connect your server

Execution Modes

MCPC provides multiple flexible execution modes to fit different scenarios:

Note: agentic, ai_sampling, and ai_acp are built-in modes — just set options.mode and they work. code_execution and code_execution_sampling require installing and loading their respective plugin packages.

Quick Example

// Interactive agent (default)
{ options: { mode: "agentic" } }

// Autonomous agent
{ options: { mode: "ai_sampling", samplingConfig: { maxIterations: 10 } } }

// Code execution with sandbox
import { createCodeExecutionPlugin } from "@mcpc/plugin-code-execution/plugin";
{
  plugins: [createCodeExecutionPlugin()],
  options: { mode: "code_execution" }
}

Detailed Documentation: See Execution Modes Guide for comprehensive information on each mode, configuration options, and best practices.

Documentation

• Getting Started - Installation and first steps
• Creating Your First Agent - Complete tutorial
• Execution Modes - Comprehensive guide to all execution modes
• CLI Usage Guide - Using the MCPC CLI
• Logging and Tracing - MCP logging and OpenTelemetry tracing
• Examples - Real-world use cases
• FAQ - Common questions and answer

Examples

See working examples in the examples directory or check out the Codex fork tutorial.

Contributing

We welcome contributions! See CONTRIBUTING.md for details.

License

MIT License - see LICENSE for details.