---
title: AI Protocol Mocking (MCP & A2A)
description: Mock MCP servers and A2A agent-to-agent protocol endpoints in MockServer to test AI integrations without calling real services.
shortTitle: AI & MCP Mocking
layout: page
pageOrder: 5
section: 'AI & MCP'
subsection: true
sitemap:
  priority: 0.8
  changefreq: 'monthly'
  lastmod: 2026-05-07T00:00:00+00:00
keywords: "mock mcp server, a2a protocol mocking, mock agent to agent, sse websocket mock"
schema_faq:
  - question: "How do I mock an MCP server with MockServer?"
    answer: "Use MockServer's SSE response support (httpSseResponse) to return Server-Sent Events from any path. Combine it with JSON-RPC response matchers to handle MCP's Streamable HTTP protocol. Your AI client under test connects to MockServer instead of a real MCP server and receives deterministic tool-list and tool-call responses."
  - question: "What is A2A protocol mocking?"
    answer: "A2A (Agent-to-Agent Protocol) is a Google-defined standard for agent interoperability. MockServer can stand up a fake A2A agent that serves the /.well-known/agent.json discovery card, handles tasks/send and tasks/get JSON-RPC calls, and returns scripted task responses. Use this to test an A2A client without running a real agent."
  - question: "Can MockServer mock SSE or WebSocket AI endpoints?"
    answer: "Yes. MockServer supports SSE responses (httpSseResponse) with per-event delays, and WebSocket mocking that performs the upgrade handshake and streams text or binary frames. Both are commonly used by LLM streaming APIs, MCP's Streamable HTTP transport, and real-time agent protocols."
  - question: "How is mocking an MCP server different from MockServer's own MCP control plane?"
    answer: "They are completely separate features. MockServer's own MCP control plane (at /mockserver/mcp) lets AI coding assistants like Claude Code or Cursor control MockServer — creating expectations, verifying calls, and debugging traffic. The AI Protocol Mocking feature lets you stand up a fake third-party MCP server that your application under test connects to. One is about controlling MockServer; the other is about mocking an external service."
---

<p>MockServer can mock AI protocol servers — including <strong>MCP (Model Context Protocol)</strong> servers and <strong>A2A (Agent-to-Agent Protocol)</strong> agents — enabling you to test AI clients against predictable mock services without running real AI infrastructure.</p>

<p>This is distinct from MockServer's built-in <a href="/mock_server/ai_mcp_setup.html">MCP control plane</a> (which lets AI assistants <em>control</em> MockServer). These features let you <em>mock other people's</em> MCP and A2A servers. For mocking LLM API responses (OpenAI, Anthropic, Gemini, etc.) with provider-correct formatting and streaming, see <a href="/mock_server/llm_response_mocking.html">LLM Response Mocking</a>.</p>

<p style="width: 60%; margin: 1em auto;"><img src="../images/mockserver_ai_protocol_mocking.png" alt="AI protocol mocking flow: MockServer receives an MCP or A2A call, matches a request expectation, and streams a scripted SSE result." style="max-width:100%;"></p>

{% include_subpage _includes/ai_protocol_mocking.html %}

<h2>Related Pages</h2>

<ul>
    <li><a href="/mock_server/ai_overview.html"><strong>AI &amp; MCP Overview</strong></a> &mdash; start here to understand all three MockServer AI modes</li>
    <li><a href="/mock_server/llm_response_mocking.html"><strong>LLM Response Mocking</strong></a> &mdash; mock OpenAI, Anthropic, Gemini, and other LLM provider APIs with provider-correct formatting and streaming</li>
    <li><a href="/mock_server/ai_mcp_setup.html"><strong>MCP Setup</strong></a> &mdash; connect an AI assistant to MockServer's own MCP control plane (the reverse of this page)</li>
</ul>

{% include_subpage _includes/ai_see_also.html %}