mcpc

MCPC (Model Context Protocol) MCPC is the SDK for building agentic MCP servers. This server enables you to compose and run MCP-powered agents by wiring together existing MCP tools and communities. With MCPC you can define multi-agent configurations, fine-tune tool descriptions and parameters, and orchestrate agents as MCP tools to create collaborative, agentic workflows. The SDK supports multiple execution modes and built-in logging/tracing, making it suitable for interactive agents, AI SDK sampling, ACP-style AI workflows, and secure code execution.

To use MCPC, install the core SDK and create an MCP server configuration that references the MCP tools you want to employ. You can run agents, load skills, and compose tools from the MCP registry. The examples show how to add a server, compose agents with tool dependencies, and run in various execution modes. Documentation highlights how to define dependencies, write agent descriptions, and instantiate the server with mcpc().

Prerequisites:

• Node.js (LTS) and npm or pnpm
• Optional: Den o, pipx, or other tooling if you plan to use alternate installation methods

Install via npm (recommended):

# Install the core MCP SDK from the npm registry
npm install @mcpc-tech/core

Install via npx (quick start):

# Add @mcpc-tech/core via jsr (if you use the JS registry) or use npm to install locally
npx jsr add @mcpc-tech/core

Deno (if you prefer Deno):

deno add jsr:@mcpc-tech/core

PNPM (alternative package manager):

# From npm registry
pnpm add @mcpc-tech/core
# From jsr registry
pnpm add jsr:@mcpc/core

Run the CLI directly without installation:

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

Tips and common considerations:

• The MCP config uses mcpServers to declare runtime commands. For node-based servers, the recommended approach is to run the MCPC core package via npx or as a local node process depending on your deployment.
• When composing agents, you can reference tools from the MCP registry and override tool parameters to suit your workflow.
• Consider using the supported execution modes (agentic, ai_sampling, ai_acp, code_execution) to match your use case. Enable logging and OpenTelemetry tracing to diagnose tool interactions.
• If you encounter environment-related issues, ensure the correct Node.js version and that your PATH includes npm binaries. Environment variables can be used to configure network endpoints, API keys, and transport settings for MCP servers.
• The README examples show how to integrate with other MCP tools (e.g., builder, skills plugins). You can replace references with your own tool set or local plugins as needed.