Skip to main content
ECP is designed to embrace and extend the Model Context Protocol (MCP). It does not replace MCP.

What MCP standardizes

  • Tool discovery — Servers expose tools with names and schemas.
  • Tool invocation — Clients call tools with structured arguments.
  • Structured tool outputs — Results are machine-readable.

What ECP adds

ECP does not redefine tool calling. Instead, it:
  • References MCP servers — Contexts declare which servers (by name) an executor can use.
  • References MCP tool names — Policies allow or deny specific tools (e.g. jira:issues.search).
  • Uses structured argument schemas — Mounts and tool calls use JSON Schema-compatible definitions.
  • Bundles permissions and policies — Default-deny tool access, budgets, write controls per executor.
  • Adds versioning and packaging — A Context is a versioned, portable manifest (like a container image).

Layered view

LayerResponsibility
MCPTool discovery and invocation
A2AAgent-to-agent communication
ECPOrchestration and execution control
If MCP is the “USB interface” for AI tools, ECP is the “container manifest” that defines the whole execution environment: which tools, which data (mounts), which guardrails (policies), and how executors coordinate.

Building with both

  • Implement or use MCP servers for your data and tools.
  • Define ECP Contexts that reference those servers, declare mounts and policies, and orchestrate agents that call MCP tools.
This keeps tool semantics in MCP and execution semantics in ECP.