mcp-tester
Instrumented MCP server for testing client conformance to list_changed notifications, enabling verification of dynamic tool, resource, and prompt surfacing.
README
mcp-tester
Instrumented MCP server for testing client conformance to */list_changed notifications and other protocol features.
Why
PR mcaps-microsoft/msx-mcp#321 proposed dynamic tool surfacing via notifications/tools/list_changed. The PR was closed with the claim that "Copilot CLI caches tools/list at session start and ignores notifications/tools/list_changed" -- but no instrumented experiment was ever run to prove or disprove this.
This repo provides the apparatus to test that claim -- and any MCP client's conformance to list_changed -- with real evidence.
What It Does
-
Probe MCP Server (
src/server.ts) -- A self-instrumenting MCP server that logs every JSON-RPC message on its stdio pipe. It exposes alpha-state tools, resources, and prompts, plus trigger tools that swap them to beta-state and send the corresponding*/list_changednotification. -
Automated Client Driver (
src/driver.ts) -- An SDK-based MCPClientthat drives the server through the full mutation cycle. This is the control: it proves the protocol works when both endpoints honor the spec. Any client that diverges from this behavior is provably non-conformant. -
Log Analyzer (
src/analyze.ts) -- Reads the JSONL log and emits a markdown evidence table answering specific questions: Did the client receive the notification? Did it re-fetch? Can it call the new item? Can it still call the removed item? -
Human Protocol (
PROTOCOL.md) -- Step-by-step instructions for running the same test against Copilot CLI or any other MCP client.
Quick Start
git clone https://github.com/ericchansen/mcp-tester.git
cd mcp-tester
npm install
npm run build
Run the automated control (SDK client)
npm run driver -- --domain all
This spawns the probe server, runs the full mutation protocol for tools, resources, and prompts, and reports results. You should see all three domains pass with full conformance.
Analyze a log
npm run analyze -- path/to/log.jsonl --domain tools
Testing Your Own MCP Client
See PROTOCOL.md for step-by-step instructions on testing Copilot CLI or any other MCP client against this probe.
Repo Layout
mcp-tester/
|-- README.md # This file
|-- PROTOCOL.md # Human-driven test procedure
|-- LICENSE # MIT
|-- package.json
|-- tsconfig.json
|-- src/
| |-- server.ts # Probe MCP server (stdio)
| |-- logger.ts # JSONL stdio tap + log writer
| |-- analyze.ts # CLI: parses log -> markdown evidence table
| +-- driver.ts # CLI: spawns server, drives it via SDK Client
+-- results/
+-- copilot-cli/
+-- .gitkeep # Committed experiment results go here
How the Probe Works
Initial State (Alpha)
| Domain | Alpha Item |
|---|---|
| Tools | mcptester_probe_alpha |
| Resources | probe://alpha |
| Prompts | mcptester_alpha_prompt |
Plus five permanent tools: mcptester_mutate_tools, mcptester_mutate_resources, mcptester_mutate_prompts, mcptester_status.
After Mutation (Beta)
When a mcptester_mutate_* tool is called:
- The alpha item is removed (via SDK
handle.remove()) - The beta item is registered
- The SDK auto-sends the corresponding
notifications/*/list_changed debouncedNotificationMethodscoalesces remove + register into a single notification
| Domain | Beta Item |
|---|---|
| Tools | mcptester_probe_beta |
| Resources | probe://beta |
| Prompts | mcptester_beta_prompt |
Logging
Every JSON-RPC message (inbound and outbound) is logged to a JSONL file. The logger taps process.stdin and process.stdout via Transform streams installed before the SDK transport, so it sees every byte without interfering with the protocol.
Limitations
- stdio transport only -- no HTTP/SSE testing yet
- Three
list_changedflows only -- future tests planned for sampling, elicitation, roots, progress, completions - Single-server testing -- tests one MCP server connection at a time
License
MIT
Recommended Servers
playwright-mcp
A Model Context Protocol server that enables LLMs to interact with web pages through structured accessibility snapshots without requiring vision models or screenshots.
Magic Component Platform (MCP)
An AI-powered tool that generates modern UI components from natural language descriptions, integrating with popular IDEs to streamline UI development workflow.
Audiense Insights MCP Server
Enables interaction with Audiense Insights accounts via the Model Context Protocol, facilitating the extraction and analysis of marketing insights and audience data including demographics, behavior, and influencer engagement.
VeyraX MCP
Single MCP tool to connect all your favorite tools: Gmail, Calendar and 40 more.
graphlit-mcp-server
The Model Context Protocol (MCP) Server enables integration between MCP clients and the Graphlit service. Ingest anything from Slack to Gmail to podcast feeds, in addition to web crawling, into a Graphlit project - and then retrieve relevant contents from the MCP client.
Kagi MCP Server
An MCP server that integrates Kagi search capabilities with Claude AI, enabling Claude to perform real-time web searches when answering questions that require up-to-date information.
E2B
Using MCP to run code via e2b.
Neon Database
MCP server for interacting with Neon Management API and databases
Exa Search
A Model Context Protocol (MCP) server lets AI assistants like Claude use the Exa AI Search API for web searches. This setup allows AI models to get real-time web information in a safe and controlled way.
Qdrant Server
This repository is an example of how to create a MCP server for Qdrant, a vector search engine.