MCPLoadTester

MCPLoadTester is a load testing toolkit for Model Context Protocol (MCP) servers.

It allows performance testing of MCP tools and agents using:

• Apache JMeter
• k6 (planned)
• CLI tools (planned)

Compatibility

Installation

This project now builds a thin plugin jar:

• plugin jar goes to JMETER_HOME/lib/ext
• runtime dependencies go to JMETER_HOME/lib

Recommended local install (plugin + dependencies)

JMETER_HOME=/path/to/apache-jmeter-5.6.3 ./gradlew installLocalWithDeps

Plugin-only local install

Use this only if dependencies are already managed elsewhere.

JMETER_HOME=/path/to/apache-jmeter-5.6.3 ./gradlew installLocal

Build artifact

./gradlew jar

Jar path:

build/libs/jmeter-mcp-sampler-<version>.jar

Plugins Manager Custom Repo (Latest Release)

Each tagged release now publishes a Plugins Manager descriptor asset:

https://github.com/AndreyVMarkelov/MCPLoadTester/releases/latest/download/plugins-repo.json

To add this custom repo in JMeter, set in user.properties:

jpgc.repo.address=https://jmeter-plugins.org/repo/;https://github.com/AndreyVMarkelov/MCPLoadTester/releases/latest/download/plugins-repo.json

Then restart JMeter and open Plugins Manager. The catalog/display name is MCPLoadTester JMeter MCP Sampler.

Quick Start (2 Minutes)

Path note: commands and scripts/... arguments below are relative to this repository root.
Run JMeter from the repo root, or use absolute paths.

1. Install plugin (see Installation > Recommended local install above).
2. Restart JMeter.
3. Add sampler: Thread Group -> Add -> Sampler -> MCP Sampler.

4. Use local echo mock (stdio) with these sampler settings:
   • Transport: stdio
   • Command: python3
   • Arguments: scripts/mock_mcp_server.py
   • MCP Method: tools/list
   • Client name: jmeter-mcp-sampler (default)
   • Client version: 1.0.0 (default)

5. Add a Response Assertion:
   • Field to Test: Response Data
   • Pattern: echo

6. Run test and verify the assertion passes.

HTTP + SSE Quick Start

1. Start local mock HTTP+SSE server:

node scripts/mock_http_sse_mcp_server.js

2. In sampler set:

• Transport: http+sse
• HTTP base URL: http://127.0.0.1:8080
• SSE endpoint path: /events
• HTTP send path: /rpc
• MCP Method: tools/list
• Validation mode: none (when using JMeter Response Assertion)

3. Add Response Assertion for HTTP status:

• Field to Test: Response Code
• Pattern: 200

4. Run test and verify the assertion passes (Response Code = 200, sample is green).

Alternative: use built-in sampler validation instead of JMeter assertions:

• Validation mode: regex
• Validation expr: echo

GUI Configuration

Screenshot guide and catalog: docs/images/README.md

Supported MCP Methods

Test MCP Servers

Local mock servers in this repo

1) Python echo server

File: scripts/mock_mcp_server.py

Start manually:

python3 scripts/mock_mcp_server.py

Behavior:

• initialize returns server mock-mcp-server.
• tools/list exposes one tool: echo.
• tools/call with tool echo echoes arguments back and returns content.
• resources/list returns one in-memory resource.
• unknown tool returns JSON-RPC error -32602.

Useful tools/call sample:

{"message":"hello","count":2}

Expected response includes:

• content[0].text like echo: hello (count=2)
• echo.message = "hello"

2) Node calc server

File: scripts/mock_calc_mcp_server.js

Start manually:

node scripts/mock_calc_mcp_server.js

Behavior:

• initialize returns server mock-calc-mcp-server-ts.
• tools/list exposes tools add and divide.
• tools/call validates numeric a and b.
• divide by zero returns error -32000.
• unknown tool returns error -32602.

Useful tools/call samples:

{"a":40,"b":2}

{"a":10,"b":0}

Expected:

• add returns numeric value (42 for sample above).
• divide with b=0 returns error Division by zero.

3) Python perf server

File: scripts/mock_perf_mcp_server.py

Start manually:

python3 scripts/mock_perf_mcp_server.py --base-delay-ms 5 --jitter-ms 10 --slow-every 20 --slow-delay-ms 200

Behavior:

• tools/list exposes sleep, payload, cpu.
• sleep tool supports controlled delay.
• payload tool returns a large blob for payload-pressure testing.
• cpu tool burns CPU loops and returns accumulator.

Useful tools/call payload samples:

{"delayMs":50}

{"sizeKb":256}

{"loops":300000}

4) Node HTTP + SSE server

File: scripts/mock_http_sse_mcp_server.js

Start manually:

node scripts/mock_http_sse_mcp_server.js

Behavior:

• SSE endpoint: /events
• JSON-RPC send endpoints: /rpc and /message
• supports initialize, tools/list, tools/call, resources/list
• sends JSON-RPC response via HTTP body and broadcasts same response over SSE

External servers often used with this sampler

mcp-server-fetch (via uvx):

Command:   uvx
Arguments: mcp-server-fetch

mcp-server-filesystem:

Command:   uvx
Arguments: mcp-server-filesystem /tmp

github-mcp-server:

Command:   github-mcp-server
Arguments: stdio

Requires GITHUB_PERSONAL_ACCESS_TOKEN.

Docker Scenarios

Each profile builds image, runs a test plan, writes JTL + HTML report to results/.

docker compose --profile mock-echo up --build
docker compose --profile mock-calc up --build
docker compose --profile perf-cold-start up --build
docker compose --profile perf-tail up --build
docker compose --profile perf-payload up --build

Reports:

• results/docker-mock-echo/report/index.html
• results/docker-mock-calc/report/index.html
• results/docker-perf-cold-start/report/index.html
• results/docker-perf-tail/report/index.html
• results/docker-perf-payload/report/index.html

Gradle Task Shortcuts

./gradlew test
./gradlew check
JMETER_HOME=/path/to/apache-jmeter-5.6.3 ./gradlew installLocalWithDeps
JMETER_HOME=/path/to/apache-jmeter-5.6.3 ./gradlew mockAll
JMETER_HOME=/path/to/apache-jmeter-5.6.3 ./gradlew perfAll

CI and Release

Additional CI/release details are documented separately.

• CI details: docs/CI.md
• Release and publishing details: docs/RELEASE.md

Troubleshooting

Results file ... is not empty in Docker:

• fixed in current entrypoint by cleaning stale results.jtl before run.

JMETER_HOME is not set:

• run Gradle tasks with JMETER_HOME=/path/to/apache-jmeter-5.6.3.

No sampler appears in JMeter:

• verify plugin jar is in lib/ext
• verify runtime dependency jars are in lib
• restart JMeter

License

MIT