MCP Rigor

Natural-language testing for Model Context Protocol servers.

Visual QA workspace · English ⇄ YAML parity · interactive evidence timeline · CLI, CI, and shareable reports — watch the MP4

MCP Rigor lets QA teams test MCP tools, resources, prompts, contracts, and transports without writing test code. Tests are deterministic: no AI model interprets the wording.

Status: 1.5.0 stable. Node.js 20 or 22 is required.

Install

MCP Rigor is published on npm as mcprigor. The package ships compiled code, ready to run:

npm install mcprigor

Or try it instantly without installing:

npx mcprigor@latest init tests/acceptance.mcpr

Use npx mcprigor in commands below, or install globally (npm install -g mcprigor) to use mcprigor directly. In a project where tests run only during development or CI, npm install --save-dev mcprigor keeps it out of production dependencies.

Related MCP server: MCP Test Server

Install from source

Only needed if you want to modify MCP Rigor itself:

git clone https://github.com/FusionOnePlatform/mcprigor.git
cd mcprigor
npm ci
npm run check   # build + full test suite
npm link        # optional: use your local build as the global `mcprigor`

Write your first test

Create calculator.mcpr:

MCP Test 1

Suite: "Calculator acceptance tests"
Server: node dist/server.js

Test: "Adding 20 and 22 gives 42"
  Call tool "add" with:
    a: 20
    b: 22

  Expect "structuredContent.sum" equals 42

Check the wording, then run it:

npx mcprigor check calculator.mcpr
npx mcprigor test calculator.mcpr

Create a browser report:

npx mcprigor test calculator.mcpr --html report.html

Expected summary:

✓ Adding 20 and 22 gives 42

1 passed, 0 failed, 0 skipped, 0 blocked

Use the QA workspace

Start the local browser interface:

npx mcprigor workspace .

Or expose MCP Rigor to AI coding agents as an MCP server (docs):

npx mcprigor serve .

Open the printed http://127.0.0.1:... URL. You can select a suite, edit it, validate the wording, run tests, run transport parity, and review results.

The workspace is loopback-only and does not accept arbitrary commands from the browser.

Connect to an MCP server

Local stdio server:

Server: node dist/server.js

Python server:

Server: python -m my_mcp_server

Streamable HTTP server:

MCP URL: https://qa.example.com/mcp

Server options:
  headers:
    Authorization: "Bearer ${env.MCP_TOKEN}"

Run with an environment variable:

MCP_TOKEN="your-token" npx mcprigor test customer.mcpr

Do not place reusable credentials directly in test files.

Common test actions

Call tool "search" with:
  query: "red shoes"

Read resource "catalog://status"

Get prompt "write_summary" with:
  topic: "Release quality"

Expect it succeeds
Expect an error
Expect "structuredContent.total" equals 2
Expect "content[0].text" contains "complete"
Expect "items" has 3 items

See the natural-language cookbook for copy-ready examples.

Compare local and deployed behavior

MCP Test 1
Suite: "Calculator parity"

Compare target "Local": node dist/server.js
Compare target "QA": https://qa.example.com/mcp

Test: "Addition is consistent"
  Call tool "add" with:
    a: 20
    b: 22
  Expect "structuredContent.sum" equals 42

npx mcprigor parity calculator-parity.mcpr

MCP Rigor runs every scenario against both targets and reports semantic differences.

Create a test from a live server

Provide a small .mcpr file containing the target, then run:

npx mcprigor author server.mcpr --out search-customer.mcpr

The guided author discovers operations, asks for inputs, previews a response, and writes a reviewable natural-language test.

Use data-driven tests

Test: "Calculator examples"
  For each row:
    | caseId | a  | b  | expected |
    | basic  | 2  | 3  | 5        |
    | larger | 20 | 22 | 42       |

  Call tool "add" with:
    a: "${row.a}"
    b: "${row.b}"

  Expect "structuredContent.sum" equals "${row.expected}"

CSV, JSON, YAML, Excel, REST, Google Sheets, SQL adapters, typed columns, filters, joins, and deterministic sampling are also supported.

Use in CI

- uses: actions/setup-node@v4
  with:
    node-version: 20
- run: npm ci
- run: npx mcprigor check tests/acceptance.mcpr
- run: npx mcprigor test tests/acceptance.mcpr --junit reports/mcp.xml --evidence .mcprigor/ci-run

Exit codes:

Main commands

mcprigor workspace [DIRECTORY]        Start the local QA interface
mcprigor init FILE                    Create an example test
mcprigor check FILE                   Validate without connecting
mcprigor test FILE                    Run natural-language or YAML tests
mcprigor author TARGET --out FILE     Build a test interactively
mcprigor parity FILE                  Compare named targets
mcprigor audit FILE                   Run deterministic security probes
mcprigor composition-check FILE       Check a mounted server fleet
mcprigor composition-discover FILE    Save a combined fleet contract
mcprigor composition-drift FILE       Gate fleet contract drift
mcprigor discover FILE                Save the live MCP contract
mcprigor contract-check LOCK --target FILE
mcprigor evidence-show DIRECTORY

Performance and governance features new in 1.5.0:

• Expect the call to finish within 800ms, percentile budgets, and --fail-on-regression

• mcprigor audit deterministic security probes with scored PDF/CSV/JSON reports

• Multi-server test routing, collision checks, combined fleet locks, and composition drift

• GitHub Action with rich PR reports

• Live MCP surface and schema coverage with --fail-under

• Scheduled HTTP monitoring with failure/recovery webhooks

• Clickable HAR-style session timeline in HTML reports

• Contract drift markdown as a downloadable Action artifact

• mcprigor publish — shareable static report URLs

See the complete CLI reference.

Editor support

The VS Code extension in editors/vscode adds syntax highlighting and inline mcprigor check diagnostics for .mcpr files. Package it with npx @vscode/vsce package from that directory, or install a release .vsix with code --install-extension.

Choose your next guide

• New QA user: Getting started

• Writing scenarios: Natural-language cookbook

• Using the browser UI: QA workspace

• Setting up targets and CI: Engineer setup

• Fixing failures: Troubleshooting

• All documentation: Documentation index

Scope and safety

MCP Rigor performs black-box behavior and contract testing. It complements MCP Inspector and official MCP Conformance; it does not provide certification.

Remote data and custom extensions are disabled unless explicitly enabled. Reports and evidence are sanitized, but business response data may still be sensitive. Review security and retention before storing CI artifacts.

Contributing

See CONTRIBUTING.md, then run:

npm ci
npm run check

Apache-2.0 licensed.