MCPC
An SDK for building agentic MCP servers by composing and fine-tuning existing Model Context Protocol tools, wrapping them into new tools, multi-agent systems, or code-execution and sandboxed sampling modes.
MCPC lets a developer build an agentic Model Context Protocol (MCP) server by describing a tool in text and referencing existing tools from the MCP community, rather than writing one from scratch; the result is a standard MCP tool that works with any MCP client. Existing tool descriptions and parameters can be modified, or their results wrapped and filtered, to adapt them to a specific business need, and each agent can itself be defined as an MCP tool so agents can be composed into larger multi-agent systems.
It supports several execution modes for a composed tool: an interactive agentic mode, AI SDK sampling, an ACP mode, secure code execution, and a combined sandbox-and-sampling mode, each with its own implementation. Built-in logging and OpenTelemetry tracing are included, and tools can declare dependencies on domain-specific skills that follow the Agent Skills specification.
It installs from npm, JSR or Deno, and a companion web site and an interactive builder agent can generate a ready-to-use MCP server without writing code directly.